Introduction to Cloud

Computing

Business & Technology
















Timothy Chou






























© 2010 by Active Book Press
2
nd
Edition
All rights reserved.

lSBN 978-0-6152-1359-0

Printed in the United States of America

i






Contents

ACKNOWLEDGEMENTS.......................................................................... V
INTRODUCTION.................................................................................... VIII
SEVEN BUSINESS MODELS......................................................................1
MODEL ONE: TRADITIONAL.........................................................................3
MODEL TWO: OPEN SOURCE........................................................................5
MODEL THREE: OUTSOURCING. ...................................................................7
MODEL FOUR: HYBRID. ...............................................................................9
MODEL FIVE: HYBRID+ .............................................................................14
MODEL SIX: SOFTWARE AS A SERVICE ......................................................15
MODEL SEVEN: INTERNET .........................................................................17
COST OF SOFTWARE...................................................................................18
SUMMARY..................................................................................................22
APPLICATION CLOUD SERVICES .......................................................24
COLLABORATION.......................................................................................25
WEBEX STORY.............................................................................................26
FINANCIAL APPLICATIONS .........................................................................28
CONCUR STORY...........................................................................................29
NETSUITE STORY.........................................................................................30
HUMAN RESOURCE APPLICATIONS ............................................................31
KENEXA STORY............................................................................................32
TALEO STORY..............................................................................................33
SUCCESSFACTORS STORY.............................................................................35
CRM FOR MARKETING ..............................................................................36
VOCUS STORY .............................................................................................37
OMNITURE STORY .......................................................................................38
CONSTANT CONTACT STORY ........................................................................40
CRM FOR SALES........................................................................................42
SALESFORCE.COM STORY.............................................................................42
CRM FOR SERVICE ....................................................................................43
RIGHTNOW TECHNOLOGIES STORY..............................................................45
VERTICAL APPLICATIONS ..........................................................................48
DEALERTRACK STORY .................................................................................51
BLACKBAUD STORY .....................................................................................52

ii
OPENTABLE STORY......................................................................................53
CLOUD SERVICE STACK.............................................................................54
SUMMARY..................................................................................................56
DATA CENTER...........................................................................................57
NETWORK CLOUD SERVICES......................................................................58
NTT STORY .................................................................................................61
POWER QUALITY........................................................................................63
POWER UTILIZATION EFFICIENCY..............................................................64
POWER CARBON FOOTPRINT......................................................................67
PG&E STORY .............................................................................................69
AVAILABILITY MANAGEMENT ...................................................................69
SECURITY MANAGEMENT ..........................................................................70
AT&T STORY ..............................................................................................71
ECONOMICS................................................................................................72
SUMMARY..................................................................................................74
COMPUTE & STORAGE CLOUD SERVICES ......................................75
COMPUTE & STORAGE CLOUD SERVICES...................................................76
VIRTUAL MACHINE....................................................................................80
SECURITY MANAGEMENT ..........................................................................82
TD BANKNORTH STORY ...............................................................................83
PUBLIC AND PRIVATE CLOUDS...................................................................83
IBM STORY .................................................................................................84
HIGH GROWTH APPLICATIONS...................................................................86
PEAKY APPLICATIONS................................................................................87
PARALLEL APPLICATIONS ..........................................................................88
HP STORY...................................................................................................90
TODAY’S CLOUDS ARE SMALL...................................................................91
COMPUTER CARBON FOOTPRINT................................................................91
SUMMARY..................................................................................................95
PLATFORM SERVICES FOR SOFTWARE DEVELOPMENT...........97
APPLICATION SPECTRUM ...........................................................................98
MULTI-TENANCY......................................................................................101
SOFTWARE DEVELOPMENT LIFECYCLE....................................................104
HORIZONTAL PLATFORM SERVICES .........................................................106
MICROSOFT AZURE STORY.........................................................................110
GOOGLE APPENGINE STORY......................................................................111
VERTICAL PLATFORM SERVICES ..............................................................113
SALESFORCE.COM STORY...........................................................................113
FACEBOOK STORY .....................................................................................117
NETSUITE STORY.......................................................................................120
TESTING...................................................................................................122
CUSTOMIZATION......................................................................................125
TRADITIONAL VS. CLOUD DEVELOPMENT................................................129

iii
SUMMARY................................................................................................131
PLATFORM SERVICES FOR OPERATIONS .....................................133
CHANGE MANAGEMENT ..........................................................................137
EBAY STORY..............................................................................................139
AVAILABILITY MANAGEMENT.................................................................142
PERFORMANCE MANAGEMENT ................................................................145
SECURITY MANAGEMENT ........................................................................146
CARE REHAB STORY ..................................................................................148
MEXICO STORY .........................................................................................149
CUSTOMER SERVICE ................................................................................150
SEARCH-BASED APPLICATIONS................................................................151
SUMMARY................................................................................................155
MARKETING............................................................................................156
ECONOMICS OF SALES & MARKETING .....................................................157
TRADITIONAL SALES & MARKETING .......................................................158
EMAIL MARKETING..................................................................................160
E-LOAN STORY ..........................................................................................161
SEARCH MARKETING...............................................................................162
SOCIAL MEDIA MARKETING ....................................................................163
GET NOTORIOUS ......................................................................................165
EDUCATE & SELECT.................................................................................167
SUMMARY................................................................................................168
SALES.........................................................................................................170
TRANSFORMING TRADITIONAL SOFTWARE SALES...................................171
SELLING NEW APPLICATION CLOUD SERVICES........................................173
CONTRACTS .............................................................................................178
SERVICE LEVEL AGREEMENTS.................................................................179
SALES COMPENSATION ............................................................................182
INSTALLED BASE SALES...........................................................................184
INDIRECT CHANNELS ...............................................................................186
FINANCE ...................................................................................................190
REVENUE RECOGNITION ..........................................................................191
MONTHLY RECURRING REVENUE (MRR) ................................................193
CASH........................................................................................................194
COST OF SALES: WHEN TO HIRE FOR MORE SALES..................................194
CHURN: COST OF KEEPING A CUSTOMER.................................................196
FINANCIAL SYSTEMS................................................................................197
FINANCIAL ANALYSTS .............................................................................200
VENTURE CAPITAL...................................................................................200
SUMMARY................................................................................................202
HUMAN RESOURCES.............................................................................203

iv
ECONOMICS..............................................................................................204
RIGHT PEOPLE, RIGHT STAGE ..................................................................204
RIGHT PEOPLE..........................................................................................207
CULTURE..................................................................................................208
SCRAPE ....................................................................................................210
BUY NEW.................................................................................................212
REMODEL.................................................................................................215
WORLD OF WORK ....................................................................................217
SUMMARY................................................................................................219
BEGINNING OF SOFTWARE................................................................221
INDEX.........................................................................................................226



v






Acknowledgements



For almost fifteen years I had the pleasure of teaching
introductory computer architecture to over a thousand students at
Stanford University. So in 2005 when I left my role as the
President of Oracle On Demand, with the help of Claire Stager, I
launched the first seminar class on software as a service. It was
born of my desire to help both technology students understand
more about the world of business as well as to have business
students understand more about technology. We just completed
the 5
th
year of the class. With nearly 50 unique guest lecturers
we have spanned the range of companies that are either
delivering their applications as a cloud service or providing the
platform to do so.

Some of these lectures have been captured explicitly and
implicitly in this textbook, so I’d like to thank many of them for
their contributions. At the application level nearly all of the
companies who’ve developed business application cloud
services have lectured including: Evan Goldberg, Zach Nelson,

vi
Marc Benioff, Subrah Iyar, Marc Chardon, Steve Singh, Greg
Gianforte, Mark O’Neil, Rudy Karsan, and Michael Gregoire.
The class also focused on leaders on the consumer side as well
including: Tony Hsieh, Samir Aora, Vic Gundotra, Udi Manber,
Reid Hoffman, Joe Kennedy, Jeff Jordan, Maynard Webb, Dave
Girourard and Philip Rosedale. At the platform and
infrastructure level the lecturers included: Charles Phillips, Brian
Behlendorf, Tom Leighton, Jonathan Schwartz, Pradeep Sindhu,
Werner Vogels, and Erich Clementi.

Finally special thanks to Professor Qianchuan Zhao, for
inviting me to deliver 10 lectures at Tsinghua University in the
winter of 2009. It was the preparation for those lectures that
created the first generation of this textbook. I’d also like to
acknowledge Janet Stevenson, for her artwork, and the early
readers of the textbook: xxx. The Chinese character on the cover
is the word for cloud. My mother did that calligraphy at the age
of ninety. We should all be so strong. For this and more I am
indebted to my parents, Mary Ann and David, whose hard work
and perseverance provided me the means to realize many of my
dreams. And finally thanks to my wife, Sue, and our three
children, Danielle, Alexandra and Caroline, without whom the
past twenty years would not have been anywhere near as
meaningful or as exciting.



vii






viii
Introduction


Cloud computing, managed services, software as a service
(SaaS), software on demand, Software+Service, platform as a
service (PaaS), infrastructure as a service have all been used to
describe new ways to build, deliver, and purchase software. Are
they just different names for the same thing, or are they similar
names for different ideas? This textbook is written as a primer
for anyone trying to make sense of what many believe is the 3
rd

major wave of computing, after mainframe, and client-server
computing. While it contains some aspects of technology the
book is first and foremost written from a business perspective.
Hence, we open with seven business models for software and
use these models in the discussions throughout the remaining
chapters. The book is written for any business contemplating
developing new application cloud services for internal or
external usage, or for those considering moving existing
applications to the cloud. In that light we introduce a five-layer
cloud services stack to educate you on many of the new cloud
services you can buy and not build. Finally, this textbook is not
written for the experts in marketing to read the marketing
section, or experts in operations to read the operations chapter.
Instead, our mission is to deliver a holistic and balanced view
useful for the technology student to understand the business

ix
challenges and the business student to understand the technology
challenges.

The origins of the book can be traced to early 2002. In March
of that year, the Managing Director of Oracle Germany asked to
meet with me as the President of the Oracle On Demand
business. During the meeting he told me he had assigned two
salespeople to selling Oracle On Demand. I thought to myself
that, of course, the good news was that we have two new sales
people, but what I couldn’t get out of my head was “what are
they saying to customers?” I realized that the way we educate
salespeople is to bring them in to the corporate campfire, tell
them lots of great stories, and send them back out into the
wilderness, hoping they will be able to repeat the stories.

Any student of history knows there was a massive change
when Gutenberg invented the printing press. Suddenly,
knowledge could stop being just campfire stories that were
repeated, abused and often lost, but rather could be written
down, printed and passed out to many people. So, over a long
weekend, I hammered out the story I had been telling around the
campfire with the intent that, perhaps during a long plane ride or
over a few evenings, an Oracle salesperson or a potential
customer could come to understand why moving to “software as
a service” was a logical and inevitable step. That book was
called The End of Software. Part of the book was devoted to

x
case studies of three small private companies: Salesforce.com,
RightNow Technologies, and NetSuite. None of these
companies was public at that time. Salesforce.com had
generated only $5M in revenue. Most people were debating why
ASPs (Application Service Providers) failed and whether SaaS
(Software as a Service) would be successful.

Eight years later, the world is completely different. To get an
idea of how different, consider since 1999 fifteen companies,
who all deliver business application cloud services, have become
public companies. These fifteen include Concur (1999), Webex
(2002), Kintera (2003), Salesforce.com (2004), RightNow
Technologies (2004), Websidestory (2004) Kenexa (2005),
Taleo (2005), DealerTrack (2005), Vocus (2005), Omniture
(2006), Constant Contact (2007), Successfactors (2007),
Netsuite (2007), and Opentable (2009).

Some of these companies have been acquisition targets.
Webex was acquired by Cisco in March 2007 for $3.2B.
Blackbaud purchased Kintera in 2008. Omniture acquired
Websidestory and most recently Omniture itself was acquired by
Adobe in October 2009 for $1.8B. As 2009 comes to a close the
remaining eleven companies have a combined market cap of
over $15B. So for those of you who wonder if companies will
purchase business application software as a cloud service I think
you have your answer.

xi

Chapter 1, “Seven Business Models” was originally published
by the SIIA
1
(Software and Information Industry Association)
and based on a talk Scott Russell invited me to give at
Diamondhead Ventures in 2005. In this chapter, we outline the
seven business models that encompass the entire software
industry, from traditional software to open source, to
outsourcing, to SaaS, to the consumer Internet. The intent is not
to argue that one model is better than another, but rather to make
sure, as a software business, you know which model you’re
trying to succeed in.

Anyone contemplating building new applications or
transforming their existing applications to cloud services can
today take advantage of 1000s of cloud service provided by
companies large and small. We introduce the idea of a five-layer
cloud service stack and dedicate Chapters 2 through Chapter 5 to
covering each of the major layers. The cloud service stack is
meant to further differentiate each layer of cloud services, each
with a different audience. Chapter 2 begins at the application
layer with application cloud services. This layer of the stack is of
interest to the business users. We’ll discuss all of the application
areas that have benefited from being delivered as a cloud service
– both horizontal and vertical. Chapter 3, “Data Center” begins

1
Chou, Timothy. Seven Vices or Seven Virtues. Upgrade Magazine, June/July 2005.

xii
at the bottom of the stack. All of the current generation of
application cloud services have consumed network and co-
location cloud services. The co-location or data center layer
focuses on delivering high quality, security and reliability
locations to house computers and storage. At $1,000 per square
foot to build these new data centers require a minimum outlay of
$100M. While not a semiconductor fabrication line, the
investment is not chicken feed. This layer is of interest to the
operations team at any application, platform, or compute &
storage cloud service business.

The next layer of the stack: compute and storage cloud
services is discussed in Chapter 4. Pioneered by Jeff Bezos and
Amazon the ability to purchase compute by the hour and storage
by the month is creating an environment for a whole new range
of applications. While having compute and storage available on
demand is important it’s likely the cloud computing war will be
won by winning the developer. In Chapter 5, “Platform Services
for Development” we discuss horizontal and vertical platform
cloud services. Horizontal services are designed to provide
many of the development environments we’ve seen in the
traditional world (e.g., Visual Studio, Ruby on Rails). However,
instead of focusing only on the developer, these new platform
services include much of the operational environments that were
traditionally separate. Vertical platform services, while more
restrictive allow the developer to leverage the data models

xiii
supported by applications like salesforce.com, Netsuite and
Facebook. Developing software in the newer models is not
exactly the same as the traditional methods; we’ll talk about how
it’s different and why it addresses some of the fundamental
challenges we’ve encountered in traditional software
development.

Chapter 6, “Platform Services for Operations” introduces
perhaps the most challenging area for traditional software
people, the management, and operations of the software. In this
chapter, we’ll address some of the major “buts”, such as “…but,
it’s not secure,” or “…but, how do we customize?” The chapter
includes an interesting discussion with Lynn Reedy, who put in
place the basic machinery that powers eBay.

After building the next great cloud service you’ll be faced
with the traditional challenge of how to market and sell the
service. Chapters 7 and 8 are written for the non-sales and
marketing professional. As the price points for these cloud
services decrease, the challenges are even greater to find new
ways to educate the buyers of your services, quickly and
inexpensively. In Chapter 7 we cover some of the traditional
marketing approaches as well as discuss some of the more
innovative solutions. Whether you’re trying to leverage your
existing sales teams or build a new sales team, you’ll need to
figure out how to make sales & marketing a more efficient and

xiv
powerful organization. All of the public application cloud
service companies spend far more on sales and marketing than
any other functional area. We’ll hear from many experts in this
area, including sales executives from WebEx, Google,
Salesforce.com, and Oracle.

In the end though, any technology company requires two
things to be successful: money and people. The chapter entitled
“Finance” is meant for the non-finance professional and
highlights some of the key financial metrics (MRR, Churn Rate,
and Customer Lifetime Value) that are important to running a
subscription business. Finally, Chapter 10, “Human Resources,”
may be at the end of the book, but we all should remember that
every software business is a people business. Having a solid
culture, stage-appropriate leadership and the right specialist in
the right organizational structure are as key as picking the right
development methodology or sales compensation structure.

I hope this book will serve to inform you, whether you’re a
business student trying to make sense of these new technologies
and business models or a technology student trying to
understand what it means to launch a new business. We believe
the campfire stories and lessons apply to all businesses. As long
time venture capitalist Ann Winblad said, “This is one thing
that's really changing in the marketplace. Everybody is a

xv
software company, not just the companies in the software
industry.”











1

SEVEN BUSINESS MODELS



In the 6th century, Pope Gregory the Great, one of only two
Popes to be given the title, “the Great”, defined the Seven
Deadly Sins, or as they are sometimes called, the Seven Deadly
Vices. These were: Pride, Envy, Gluttony, Lust, Anger, Greed
and Sloth. Not only did Pope Gregory define the seven sins to
avoid, he also included a balancing set of values, the Seven
Virtues: Faith, Hope, Charity, Fortitude, Justice, Prudence and
Temperance. Many would agree that managing a software
company requires a little of both.

CLOUD



2
Today, as the software industry undergoes a major
transformation to a cloud computing model, it’s important to
understand the seven potential business models before deciding
if they are a Vice or a Virtue. You will hear some say, “We
can’t move to a SaaS model, the sales guys will never be able to
sell it,” while others will talk endlessly about the virtues of the
model and how, “It’ll be great to not have to wonder if this
quarter we’ll close that big deal at midnight on December 31.”
As you’ll see some companies operate in only one model; others
have products that operate in multiple models.

The seven business models are complete and include every
company that makes money from their software. There are
successful companies in each of the models. The models are
oriented around application software and we’ll start from the
traditional world of business software.

Figure 1.1 Seven Business Models

SEVEN VICES, SEVEN VIRTUES



3
Model One: Traditional
This is the traditional software business model, mastered by
many companies, including Oracle and SAP. The software is
licensed in perpetuity with a one-time price, in our example, of
$4,000 per user. Subsequently, the customer is charged some
percentage; say 20 percent, for the update/upgrade rights as well
as maintenance, sometimes referred to as the software support
fee. In our case study, this translates to about $65 per user per
month, or around $800 per user per year. Many would point out
that even traditional software companies have large subscription
businesses. In fact in late 2009, Oracle’s maintenance business is
over $12B annually.

The business that bought the software then buys hardware,
and hires or redeploys people to manage the software. It’s here
that the spending really goes up. The Gartner Group has
estimated that the end user can spend up to four times the cost of
their software license per year to manage these applications.
This cost multiplier is not limited to complex business software.
Why does it cost so much?

Amy Konary, at International Data Corporation (IDC), and
one of the earliest industry analysts to follow the change to
software delivered as a service, cites five major areas that cost
CLOUD



4
companies millions of dollars annually. “CIOs and their
departments really focus on five key aspects of the management
and maintenance of their computer and software systems:
availability, performance, security, problems, and change
management.” The implications of this are substantial. First,
CIOs’ budgets are dominated by spending on managing
software, leaving little money on the table for new programs and
projects. In our example, while the purchase price of the
software may be $4,000 per user, the customer will end up
spending $1,300 per user per month just to manage that
software.

Some software companies also provide additional support
services, so called premium support, which often conflict with
their partners or internal consulting groups who also provide
value added services to help the customer service the software.
A Model One business is good since, with the large up-front
purchase of the software, sales and marketing expense, which
typically can run thirty to forty percent of revenue, can be
funded. At a well-run Model One business, you’ll see license
revenue breaking even, and all of the profit realized from
software support. Software support or maintenance is mostly
limited to break-fix; access to a Web site of usage information
and upgrades. It is, of course, in the software company’s interest
SEVEN VICES, SEVEN VIRTUES



5
to move customers forward on new releases so as to not incur
expense in supporting multiple releases of the software. In the
early stages of any software company, this is mutually
beneficial, as the customer needs and wants future functionality
and, in many ways, that’s why the consumer chooses to buy
versus build. WebEx founder Subrah Iyar, lecturing at Stanford
in 2006, said “I would much rather have built a software
business on the traditional model—it’s a much better business
model for the software company.”

Model Two: Open Source
The free software movement began in the late 80s. By 1998,
a group of programmers came up with open source as a
replacement for free software, creating the Open Source
Initiative (OSI) to promote the new term. Under the name "open
source," an in-flux of for-profit dollars has changed the
landscape of free software remarkably. Most major free
software projects now have corporate backing of one form or
another, and many of these companies have become quite adept
at working with the community that participates in the projects,
and the users who report bugs and suggest new features.

In May 2005, IBM made an acquisition of Gluecode
Software, a privately held company specializing in integration
CLOUD



6
software based in El Segundo, CA. The purchase price was
estimated at $100M. A year later, in June 2006, Red Hat, the
world's best-known Linux distributor, announced the acquisition
of JBoss for a reported $420M. In early 2008 Sun Microsystems
surprised everyone with their $1B acquisition of MySQL.

All of this has not been lost on the investment community.
Over the past many years, at least $500M has been invested in a
variety of open source software companies. Why? The open
source model (much like Google) has reduced the entry cost to
“try it” to zero. MySQL estimates there have been 100 million
installations and over 60 thousand downloads per day.
JasperSoft estimates there have been nearly two million
downloads and over 20 thousand corporate deployments in 200+
countries of their Jasper Reports. John Roberts, the founder of
SugarCRM has shown a Google map with the heartbeat from
hundreds of computers in Europe running SugarCRM
software—and SugarCRM has no sales force in Europe.

While having people download, install, and potentially use
your software is far better than having them download a
brochure, open source software companies have struggled with
the business model. Support and maintenance, which has been
the lifeblood of Model One software companies, has been
SEVEN VICES, SEVEN VIRTUES



7
difficult to monetize. Traditional bronze, silver, and gold
support has also been difficult to sell, since there has been little
value-add other than changing the hours you will answer the
phone and how quickly you’ll respond to a request. Some open
source companies are experimenting with providing both
appliances and an on-demand solution as a way of providing a
higher degree of service than traditional software support. Some
are providing additional functionality in a paid-for subscription
service not available in their free version. Unfortunately, other
than Red Hat with their Red Hat Network, revenue has been
elusive. But with lots of money and smart people, it may not be
for long.

Model Three: Outsourcing.
Recognizing that the cost to manage software systems is
much larger than the purchase price of the software, companies
have emerged who are eager to take on the challenge. EDS
(purchased by HP) and IBM Global Services have each built
large, multi-billion dollar businesses outsourcing the
management of the software and systems. In Model Three
software is still licensed from the software company on a
perpetual basis; support fees are still paid, but now the Model
Three business will take on the management of that software for
less than the $1,300 per user per month the customer would be
CLOUD



8
spending. Outsourcers can manage the software either at the
client’s site or at their own locations. As much of the cost of the
software management is buried in the cost of people, a new
group of companies, such as Infosys, have emerged, using
offshore resources in India to lower the overall cost. Of course
today every Model Three business is leveraging a global pool of
talent whether in India, China or Eastern Europe.

Model Three can grow big companies and high value
companies. EDS at the time it was acquired was over $20B in
revenue. IBM Global Services generates over 50 percent of
IBM’s revenue, which, in 2007, was $50B. Tiny Infosys with
over $3B in revenue has traded at more than eight times revenue,
a multiple most software companies would be glad to have.
These companies continue to grow and take on higher value
functions, including entire business processes. ADP is probably
the granddaddy of this business model.

While Model Three is good, there are two key challenges.
First, over time, massive differences in labor rates are not
sustainable. Second, the dependence of system availability on
people, with the inherent risk of human error, is also not
sustainable. IBM has recognized that they could grow their
Global Services business by a factor of two, bringing them to
SEVEN VICES, SEVEN VIRTUES



9
over $100B in revenue. While the opportunity is there, they
would erode their margins. Fortunately, or unfortunately, they
are already in the low cost countries—Brazil, Russia, India, and
China—so unless there are highly skilled resources in Antarctica
waiting to be discovered, IBM will need to invest in technology.
Jim Spohrer founded an initiative inside IBM called Service
Science. Spohrer has said, “One grand research challenge for us
is how to scale up services and how to invest to get year-over-
year improvements.” Computer science has benefited from
Moore’s Law (where the capabilities of computer chips double
about every 18 months). Could a Moore’s Law of service
science be possible?

Model Four: Hybrid.
For most traditional software companies, Model Four, the
Hybrid model, represents a bridge from Model One to Model
Six: This business model attempts to leverage what an existing
software company already has: products, distribution channel,
and an installed base. In Model Four the traditional licensing
model is retained. In our example, that means the software is
still licensed for $4,000 per user with a support fee of 20 percent
or $65 per user per month. The big difference in Model Four is
to give the purchaser the choice of having the software company
service their own software. This is the model Oracle pursued
CLOUD



10
beginning in 1999 and resulted in the fastest growing business
inside of Oracle in 2004. Customers ranged from middle market
to Fortune 500 companies and spanned the range of applications
from financials to HR, manufacturing, and CRM and supply
chain. From the consumer’s perspective, the choice was
between servicing the software themselves for around $1,300 per
user per month, having a Model Three business provide the
service for less than $1,300, or having Oracle service their own
software for $150 a month. We’ll come back to answer the
question as to why was it so inexpensive, but first let’s make
sure you’re clear on the model.

In Model Four, the purchase of the software and support
remain the same. What’s new is the customer is given an option
to have the availability, security, performance, and changes
managed for an additional fee. With Oracle On Demand, the
customer had a one year contract that could be cancelled with
30-days notice, and rather than have a complex Service Level
Agreement (SLA), there was a “Nordstrom’s guarantee”. Oracle
guaranteed if for any reason the customer were unhappy with the
service, there would be a 20 percent rebate in that month, no
questions asked. It’s important to note this decision enabled
revenue to be recognized on a monthly basis.

SEVEN VICES, SEVEN VIRTUES



11
Unlike some lighter weight business processes (e.g., email,
sales force tracking, recruiting) customers of Oracle applications
implemented much heavier weight processes (e.g., process or
discrete manufacturing). As a results there needed to be the
flexibility to customize and integrate the applications. An
innovation was provided called CEMLIs, which isolated changes
that were made for one customer, which could be separately
priced and give a customer the choice of how much they valued
their customizations. There will be more on this in a later
section.

Additionally, customers were also given the flexibility to
choose where the computers would be located—either at an
Oracle owned data center or at a location of their choice. There
will be more on this later as well. Figure 1.1 illustrates this
flexibility by using the symbols of @H (for at home) meaning a
data center owned by the business providing the service. @C
(for at customer) means a site the customer owns.

Model Four also allowed customers to choose when changes
would be made to the application and when upgrades to major
releases would occur. While there was a great deal of flexibility
versus what you will find in either Model Six or Model Seven,
there was also a great degree of standardization. This started
CLOUD



12
from the hardware and moved up to the disaster recovery
process. For example, all applications were run on a Linux
multi-tier grid, disks were network attached versus SAN
attached; and, while you could choose when upgrades happened,
they always followed the same process.

Which gets us back to—how was it possible to offer the
service at significantly lower prices than a traditional
outsourcer? Could this business be operated at “product level”
margins? The quick answer is, yes, product level margins could
be achieved because of a high degree of standardization.
Standardization allows for specialization. Specialization allows
for repetition, the key to quality and skill in any area, whether
managing computers or doing heart surgery. With repetition
based on specialization, ultimately computers could be put to
work automating the key business processes in the business of
servicing software. A simple example is upgrades. Oracle On
Demand upgrades were done hundreds of times in a year by
specialized teams, and increasing parts of the upgrade process
were automated. Automation ultimately results in both lower
cost and higher quality service. By the way, this principle is
followed in spades by Model Six and Seven businesses.

SEVEN VICES, SEVEN VIRTUES



13
The advantage of this model is that the software does not
have to be re-engineered. And, from a sales perspective, since
the traditional license model is preserved, you can use your
existing sales channel. The model is not cannibalistic. In fact,
since the sales teams were compensated not only on the sale of
the license but also the first year of the service, the average
selling price nearly doubled for on-demand applications. In
addition, the ability to offer a complete solution from one vendor
was often a significant competitive advantage against traditional
software companies who had to go to partners for the solution.

Finally, by having a model that split the software license from
the service, the entire installed base of customers who had al-
ready bought the software was open for business. There were
many situations where there were no traditional software sales
persons assigned to customer accounts that had bought a lot of
software, because they had bought all the software they could
eat—an interesting paradox.

Of course, any model is not without its downsides. Alternate
channels of software distribution do not welcome Model Four.
They have made a living selling “boxes and bodies” and see the
vendors’ entry into this space as competitive. The second
downside is the model can only take you so far from a cost
CLOUD



14
perspective. The cost, not price, of providing this service is in
the neighborhood of $50 per user per month, which compares
favorably with doing it yourself or having an outsourcer provide
the service, which is probably closer to $500 per user per month.
But if someone could provide 80 percent of the function of your
product at $5 per user per month, then you would no longer stay
in Model Four. This of course is the story of Siebel and
Salesforce.com. More on that later.

Model Five: Hybrid+
Many people see the SaaS model as a change in the way
software is purchased—in particular, the elimination of the
upfront license fee. While Model Four allows for the delivery
model to change, it is easy to see you are not far from moving
the business model as well. In our example, if one were to
assume a 36-month time to pay off the $4,000 per user license
fee, this would equate to $110 per user per month. If you then
add the $65 per user per month for support and the $150 to
manage the software results in an “all in” price of $325 per user
per month. Of course, the revenue that is recognized in the first
year is less than in Model One, but midstream in the second year
this begins to change. By the third year, this model is resulting
in much larger revenue stream than was true in the traditional
SEVEN VICES, SEVEN VIRTUES



15
model. The choice of moving from Model Four to Model Five is
strictly a business choice, not a technology choice.

Callidus Software, a provider of sales performance
management solutions, has adopted this business model. Rather
than split the license from the service as we have seen in Model
Four, they are providing one price for the entire service. Since
Leslie Stretch became CEO in December 2006 the company’s
subscription revenue as a percentage of total revenue has nearly
doubled, representing nearly 40% of total revenue.

Model Six: Software as a Service
As we discussed briefly in Model Four, the cost to manage
the software can be driven down through standardization,
repetition, and automation, but there is a floor. If you get to start
from scratch and engineer an application for delivery as a service
(and only a service) the cost structures drop by another order of
magnitude. Most, if not all, of the new business software
companies started in the late 90s have adopted this model. As a
business model, even MMORPG (Massively Multi-player
Online Role Playing Games) like World of Warcraft have
adopted Model Six.

CLOUD



16
Software engineered for Model Six will have dramatically
lower cost structures than software as a service offered by
traditional software companies. Siebel had to face the direct
challenge of Salesforce.com, UpShot and others. While these
new companies’ offerings had far less functionality, the cost and,
as we’ll see, simplicity of purpose allowed them to compete very
effectively.

Model Six is not without its challenges. While it is clearly a
better model for the consumer of the software, one can debate
whether it’s a better model for the software company.
Unfortunately, rather than being able to close million-dollar
deals for software licenses in Model One, and being able to plow
that cash back into sales and marketing, companies in Model Six
have to find a way to finance the early acquisition of customers.
This has typically been done in the private and public markets.
If you analyze Salesforce.com, WebEx, and Concur you will
find it took somewhere between $50M and $100M to establish
the distribution channels required to achieve their meteoric
growth.

Furthermore, as in Model Four and Five, the existing
channels, such as the Microsoft channel, are at best ambivalent
to the Model Six players. With no easy distribution channel,
SEVEN VICES, SEVEN VIRTUES



17
many companies in Model Six reach $10M in revenue and have
no ability to grow larger.

Model Seven: Internet
Model Seven encompasses all of the companies in the
consumer Internet. Amazon.com, Facebook, Google, Yahoo and
eBay are all software companies. Model Seven is distinctive in a
few ways. First, these companies have chosen unique and often
indirect ways to monetize their intellectual property, the
software. Whether it is eBay charging by transaction, Google by
the ad, or Zappos by embedding it in the sales transaction, each
of the companies in Model Seven has adopted business models
that are asymmetric. Asymmetric business models mean that the
user of the software is not directly paying for the usage of the
software. We all use Google for free; it’s the advertisers that are
paying for it. Of course that does not mean Model Seven
companies have no cost of sales and marketing. You only need
to look at eBay’s or Google’s financials to realize that there is a
healthy spend on sales and marketing.

The second big difference is that each of these solutions is
highly specialized. You don’t buy books, auction cars or find
your perfect mate on Google. Google, eBay, and Amazon.com
have all been very focused on a single function. Why? Again, it
CLOUD



18
comes to cost. In the traditional world, if I built some interesting
software in Sydney, Australia, and I wanted to expand to
Melbourne, I had to hire a sales/support engineer and a
salesperson. My cost of distribution meant that I needed to
extract the greatest degree of value out of each of my customers.
That model resulted in the classical software company “going
horizontal”—meaning they began to offer more and more parts
of the solution, all the time promising full integration, so no new
company could get in the door. This has led to massive suites of
applications. Unfortunately, these applications tend to be
complex, difficult to install, and difficult to integrate—no one
needs to spend much time comparing a traditional enterprise
application to consumer Internet applications to see the
difference. Of course, this all shows up in the speed of adoption.
The simplicity of the Google interface begged for millions, if not
billions, of people to adopt—clearly not possible if the
application tried to do everything.

Cost of Software
If you went to work for Dell you'd be well educated on the
cost-or economics-of hardware, so what is the cost of software?
I'm not talking about the price of the software, but the cost.
Some will say the cost of software is the cost of the CD or the
manual, but, in fact, the cost of software is much, much more.
SEVEN VICES, SEVEN VIRTUES



19
I'm going to use eBay to help explain how the cost of traditional
software differs from that of delivering software from the cloud.

Let's assume ten years ago we wanted to start a company
building auctioning software. We would have hired a bunch of
smart programmers. They'd have just gotten out their C++
compilers and ultimately created an auction product. But the
first question the CEO would have asked would be, “Okay, how
do we get it to anybody?” Some smart VP of Sales and
Marketing would have said, “No problem, boss. I got the plan.
Let's put it in the Sunday paper,” so a million CDs would go out
in the Sunday paper. What would have happened on Monday
morning? My guess is the 1-800 number would be ringing with
calls: “Hey, wait a minute. This really ruined my Windows 95
configuration, and my ShockWave DLL doesn't run anymore.”
“No problem,” we would say, “We're software guys, and we can
solve this.” Then we'd hire a support team, set up a help desk,
and so on.

A few months or years later, some bright product manager
would show up and say, “Hey, boss, got a really great idea. I got
a new feature I want everybody to see. It's called 'Buy It Now.'
Let's put it in the next release.” No problem, right? We would
insert the CD in the next Sunday paper. Monday morning would
CLOUD



20
arrive and the first phone call would start with “My Linux server
just fell over after I installed your software.” No problem. We'll
hire a bunch more support guys, we'll move them to India.
Problem solved.

Now, all of us in the traditional software business know how
to do this. We've been doing this since the beginning of
software. The problem is economic. What would have been the
cost of that scenario? It would have been enormous. There is no
way eBay could possibly have existed under this software cost
model. The cost for delivering the software, distributing it, and
managing it would have been so prohibitive eBay would have
had to charge hundreds of dollars for every auction. While you
could have built a small auctioning software business, no way
you would have been a multi-billion dollar business.

If you think about what has transformed the entire hardware
industry, it has been a relentless pursuit of taking cost down by
an order of magnitude every decade. It has completely
transformed the hardware business. In the beginning of the 80s,
hardware companies like Digital Equipment Corporation,
Burroughs, and Data General dominated the scene and
companies like Intel were dismissed as not delivering real
computers. “Well, that microprocessor stuff is cool, but it's
SEVEN VICES, SEVEN VIRTUES



21
never going to amount to much.” We all know the end of that
story.

So what do the real numbers look like? If you take enterprise
software-say from SAP or Oracle-and you deliver it as an on-
demand service, you're going to do really well if the cost (not
price) is $50 per user per month. By the way, you're going to
have to standardize the hell out of everything, so you'd do really
well to get it out there at $50. Now remember, chances are, if
the customer is running this software in house, it's costing them
closer to $1000 per user per month, so even if SAP or Oracle
charges $150, it's significantly lower than having each and every
customer of the software solve the problems uniquely.

Now, if you become a student of Salesforce.com, and analyze
their cost (the cost of delivering the software), guess what that
number is?-about $7 per user per month. That's impressive.
Remember, of course, this is for a fixed set of functions, and if
you want to control when you upgrade, forget about it. But
that's not the end of our story. If you go down 101 to Mountain
View and estimate what it costs Google to deliver their service,
you'll come away with a number closer to $0.70 per user per
month. Bottom line: anyone in the software business needs to
become a student of the economics of software.
CLOUD



22

Summary
While it may not be clear to you which model to choose, one
thing is clear—the traditional software model is ending. With
increased competition and the dynamics of the open source
movement, the days of million-dollar software licenses are
clearly over. And, with increased pressure on top line growth
without fundamentally altering the cost structure of software
companies, we will be at the end of software.

You will have many choices. You can choose to use Model
One and Model Four, which is what Oracle did. You can choose
to move completely to Model Six, which is what Concur did.
You can choose to start out a new business in Model Six, which
is what Salesforce.com did. You can choose to acquire Model
Six companies, and have some of your product line in Model
One, some in Model Four and some in Model Six, which is what
Blackbaud did.

Whatever you choose to do you’ll need to think about how
the choices affect the entire company; since unlike the change to
client-server, this shift is much more than technology. In the
next chapter we’ll drill down a little more into some of the
SEVEN VICES, SEVEN VIRTUES



23
application areas and companies that have been successful in
delivering their applications as a cloud service.










24

APPLICATION CLOUD SERVICES


It’s far easier to talk about business models and the new cloud
computing services from the point of view of an application any
of us would use in a business setting. In this chapter we’re
going to introduce you to a variety of both horizontal (everyone
can use) and vertical (specialized to a particular industry)
applications. We’re going to focus only on application areas,
which have companies who have succeeded in bringing their
companies into the public market.

I was having lunch in early 2007 with an old friend Tom
Kucharvy. He said, “You know, Tim, you’ve been talking a lot
APPLICATION CLOUD



25
about this software as a service, software on demand thing. Now
when is this going to really happen?” His question got me to
thinking, so I did a little homework. I went back to 1999 and
looked up nine (at the time) of the business application software
companies, who all deliver their software as a service, and who
all have gone public since 1999. Since then some of them have
been acquired (Webex, Kintera, Websidestory, Omniture), but
the remaining eleven have a combined market capitalization of
over $15B, so it’s safe to say, “It’s happened!” So let’s start our
discussion of what areas have seen success.

Collaboration
Of all of the application areas, perhaps the simplest (from a
business point of view) is collaboration. The way you use email
vs. your friend at another company is not significantly different.
As a result some of the more successful application cloud
service companies have started out in collaboration.
Collaborative software or groupware includes features such as
email, calendaring, conferencing, instant messaging, wikis and
simple project management. The granddaddy of all collaborative
software, Microsoft, of course delivers email (Hotmail) but more
recently introduced Exchange Online and SharePoint Online.
Microsoft Office chief Chris Capossela predicts
2
that 50% of

2
Fast Company, October 28, 2008.
CLOUD



26
Exchange mailboxes will be online within five years. They
might be able to achieve this, as there are reports of large
companies, including Coca-Cola and Nokia, using their
collaboration cloud offerings.

This, of course, is a major battleground area. Google has
brought its considerable resources to bear by introducing Google
Apps, which includes mail, calendaring, and the ability to
collaborate around spreadsheets, presentations and text
documents. IBM, not to be left behind has introduced Lotus
Live and continues to innovate in this area. But perhaps the
company that did the most to prove that collaborative application
cloud services could be a high growth standalone business is
Webex.

Webex Story
Min Zhu and Subrah Iyar founded WebEx in late 1996 as a
re-start of another company called FutureLabs. FutureLabs
focused on providing software-client and client-server solutions
for data conferencing. In 1995, FutureLabs was acquired by
Quarterdeck. It was then that Subrah, who was responsible for
Quarterdeck’s acquisition of FutureLabs, met FutureLabs
founder Min. Within a year of the acquisition, Microsoft
introduced NetMeeting and ended Quarterdeck’s interest in the
APPLICATION CLOUD



27
space. Fortunately, Iyar and Min saw the potential of continuing
as a service provider, so they launched WebEx.

Although they started WebEx in 1996, it took Subrah and
Min until 1999 to actually launch the service. During 1997 and
1998, they spent a lot of time working with customers such as
Charles Schwab and StrataCom, which, ironically, was also
purchased by Cisco. In 1999, WebEx generated about $2.5
million in revenue with about 400 customers; by 2000, revenues
were $25 million, and the rest, as they say, is history. WebEx
went public in January 2002. By their 10th anniversary, they
were generating nearly $500M. With over two million
registered users and over 85,000 meetings per day around the
globe, continues to deliver one of the largest application cloud
services in use today.

While collaboration is the modern day telephone, perhaps the
first business software any company purchases is a financial
application. Whether an Excel spreadsheet, Intuit QuickBooks,
Microsoft Dynamics, Oracle Financials or SAP every company
large or small needs a financial application.

CLOUD



28
Financial Applications
Financial applications record and process accounting
transactions within functional modules such as accounts payable,
accounts receivable and payroll. Financial applications can help
provide cash & treasury management, financial control and
reporting, financial analytics, governance, procure-to-pay, and
travel & expense management. The most complex and expensive
business accounting software is frequently part of an extensive
suite of software often known as enterprise resource planning
(ERP) software.

This area has not been easy to penetrate for application cloud
service companies, perhaps for two reasons. First, almost every
company has a financial application and second, because that
application generally has few users. That being said two
companies have been quite successful in this area. Concur made
an initial public offering in 1999 and has grown a large business
specializing in travel and expense management. Netsuite, who
made a public offering in 2007, has mirrored the “suite-
approach” pioneered by Oracle and SAP and instead focused on
the middle market.

APPLICATION CLOUD



29
Concur Story
While Oracle was one of the leaders in moving from a
traditional model to software delivered as a service, few
examples compare with Concur. Concur began life as a
traditional software company, but in June 2000, chairman and
CEO Steve Singh made a bold decision to reinvent the whole
company and make the transition to delivering their expense
management software as a service. This was a major challenge,
particularly since they were a public company. The stock sank
to below $1 a share, and the number of employees went from
700 to 300. By the end of the transformation, nearly the entire
senior management team was replaced. Of course if you were
smart enough to purchase the stock at the low point your returns
would be over 40x today.

Since those lean years early in this decade, Concur has
steadily held to its mission. By the end of 2006, Concur had
over 4,000 clients, with seven of the world's ten largest
companies using their services to help them manage the
corporate travel and expense process. Their customers include
Vivek Ahuja, Global Operating Manager at Agilent, a global
measurement equipment company; Courtney McCoy, Senior
Business Systems Analyst at Cummins, Inc., a leader in diesel
engine technology and Gloria Brown, A/P & Payroll Supervisor
CLOUD



30
at Stryker Medical, a $5B+ medical technology company. While
some question whether such a transition from a Model One
provider to a leading Model Six service provider would be
possible in the public markets today, few can debate the success
of the bold decisions that Steve Singh and Concur made.

Netsuite Story
NetSuite is a leading provider of integrated online software
that allows SMBs to manage their entire business in a single
application. Evan Goldberg, Dave Lipscomb, Bill Ford, and
Chris Blum founded the company under the name NetLedger in
October 1998, with financing from Larry Ellison. Today, Evan
Goldberg is Chairman of the Board and Zach Nelson is the
company's CEO. The company launched its first online
financial application targeting small businesses in 1999. With
this foundation, the company broadened the application to
manage both back-office and front-office sales, support and
marketing. The company changed its name in September 2003
from NetLedger to NetSuite and made a successful IPO in
December of 2007.

Netsuite’s customers include Todd Spartz, CFO, Nomis
Solutions, a price optimization solution for the financial services
industry; Everette Phillips, President and CEO of China
APPLICATION CLOUD



31
Manufacturing Network and David Stover, CFO of Asahi Kasei
Spandex America. Asahi Kasei, located in Charleston, S.C., is a
subsidiary of Tokyo-based Asahi Kasei Fibers Corp., a Tokyo-
based manufacturer of polyelastane filament fibers used mainly
in textiles. You might find it hard to believe but they replaced
their SAP R/3 with Netsuite in 2008.

Human Resource Applications
Along with financial applications, human resource (HR)
applications were an early example of widely adopted business
applications. Peoplesoft, is perhaps the best known company
specializing in HR software. It was founded in 1987 and
ultimately acquired by Oracle for $10B in 2004. Human
resource applications are built to assess, acquire, develop, and
align a businesses workforce. Beginning with tracking of
resumes of prospective employees to keeping record of
employees’ compensation & benefits, performance appraisals,
HR software tracks the lifecycle of many companies most
valuable asset: their people.

It turns out this has also been a good area for application
cloud service companies including three companies that have
made initial public offerings in the past five years – Kenexa in
June 2005, Taleo in October of 2005 and Successfactors in 2007.
CLOUD



32

Kenexa Story
While Salesforce.com and WebEx started out knowing what
they wanted to be, Kenexa took the long way home. Kenexa is
the oldest of the group, having been founded by Rudy Karsan in
1987. Kenexa began operations in 1987 not as a software
company, but as a recruiting firm. Over the past 20+ years,
Kenexa has bought and sold over twenty businesses. Through
these acquisitions, Kenexa has grown its business to over $160M
in revenue in 2009. Today, Kenexa is a provider of software:
proprietary content, services and process outsourcing that enable
organizations to more effectively recruit and retain employees.
But Kenexa’s path has not been without its challenges. At a
lecture at Stanford University, Karsan talked about a day in 1991
when he had mortgaged his house, maxed out his credit cards,
and was three months late in the office rental payments. On a
certain Friday, the landlord had told him the sheriff would be
coming on Monday to evict them. The good news was that the
company had sold a $40,000 contract to a large insurance
company. Typically, the customer would have waited to the last
day to pay, but for some reason the checked arrived early, on
that Friday. Karsan took the check, drove from Philadelphia to
New York City, cashed the check at the issuing bank, and was
ready on Monday morning with his rent payment. His message
APPLICATION CLOUD



33
to the students was, you might be as hard working and as smart
as the next group, but, in the end, success is based on a lot of
luck. This was his lucky story.

Taleo Story
Taleo traces its beginnings to 1996, when founder Martin
Ouellet created Viasite, which would become the most
successful job board in Canada. Following on that success,
Ouellet founded Recruitsoft in 1998 and joined forces with
Louis Tetu, who had a background in supply chain management.
Tetu understood the value that, by leveraging the power and
connectivity of the Internet, one could deliver a talent supply
chain management solution that could provide the same bottom-
line and process improvement benefits to large enterprises that
supply chain management had brought to manufacturing.

Taleo joined fellow application cloud service companies
DealerTrack, Kenexa and Vocus in going public in 2005. By
late 2007, Taleo had over 12 hundred customers, one million
users, and had tracked over 66 million candidates. Taleo’s
software is used by Fortune 500 companies as well as smaller
companies to match all sources of talent—be they professional
and hourly candidates, agency referrals, campus recruits,
contingent workers, or existing employees—to all positions,
CLOUD



34
whether it is centralized, decentralized, or multinational. Some
of Taleo’s customers include Randy Goldberg, VP of Recruiting
at Hyatt Hotels & Resort, which manages over 100,000 rooms in
over 40 countries; Christoph Thoma, Head of HR at Roche, a
global healthcare company employing over 75,000 people
worldwide; and Bruce Hatz, Global Staffing Director at Applied
Materials, an $8B+ provider of semiconductor fabrication
equipment.

CEO Michael Gregoire is uniquely qualified to see the
similarities and differences between traditional and application
cloud service companies, as he was an Executive Vice President
at PeopleSoft before joining Taleo. He sees the challenges of
the traditional software business as being interlocked with the
technology. Meaning that, today, if a traditional software
company wants to deliver any significant function, there must be
a schema change. With a schema change comes an upgrade,
and, given the often-high degree of customization most
customers apply, the cost for the upgrade becomes prohibitive.
Hence, the business can’t move forward since the technology
can’t move forward. Mike’s lecture at Stanford in 2007 was
entitled: “Does Business Evolution Drive Technical Innovation
or Does Technical Innovation Drive Business Evolution?” He
ended the talk by saying, “Yes!”
APPLICATION CLOUD



35

Successfactors Story
Sometimes you can tell a company is different when you
walk in the door. A few years back I went to the SuccessFactors
headquarters to meet with Lars Dalgaard, founder and CEO.
The office had no cubicles, no separate offices, and furniture that
had been specially designed to reinforce that each individual was
part of a team. Many companies are born on the personal
experiences and frustrations of their founder. Successfactors is
no different. On graduation Lars went to work for Novartis, a
large pharmaceutical company based in Switzerland. In short
order he was headhunted by Unilever. When he went to the VP
of HR to inform him of his decision the VP got a key out of his
pocket, opened a closet and pulled out a huge three ring binder,
skimmed through it, and then pulled out all a set of org charts
and said, “See, here you are. You are going to be running
Portugal in two years!”

Lars said: “What? What are you talking about?” The VP
came back and said, “Well, that’s not the only one. I also have
you in Germany, but that is a lower job. I also have you …” and
he started going through more and more org charts. All Lars
could think was “Is this is how they manage careers?”

CLOUD



36
Out of that epiphany he founded Successfactors in 2001. In
2007 they made a successful initial public offering, and in 2008
placed number 133 on Deloitte’s 2008 Technology Fast 500.
Their solution is deployed to over four million end users in over
30 languages. Their customers are as small as Jay Greenstein,
owner of the Sport and Spine Rehab centers based in Virginia
and as large as Brenda Leadley, a VP of global human resources
for Allianz Group. Allianz Group was founded in 1890, and is
one of the leading global services providers in insurance,
banking and asset management with more than 180,000
employees worldwide.

CRM for Marketing
Customer Relationship Management (CRM) is a company-
wide business strategy designed to reduce costs and increase
profitability by managing a customer lifecycle from initially
marketing to find and educate the customer; ultimately close
business with the customer, service the customer and ultimately
deliver more products and service to that customer. CRM’s goal
is to bring together information from all data sources within an
organization (and where appropriate, from outside the
organization) to give one view of the customer. The goal is to
allow customer-facing employees in sales, customer support, and
marketing to make quick, informed decisions.

APPLICATION CLOUD



37
This area has been very fruitful for application cloud service
providers. In the next three sections we’ll divide up CRM into
CRM for marketing, sales and finally service. CRM for
marketing includes solutions to track PR effectiveness, email
marketing, web analytics and soon new innovations in social
media marketing. If the software can increase the number of
qualified leads for sales then it’s CRM for marketing. This group
includes products from Vocus, Omniture and Constant Contact.

Vocus Story
Vocus was co-founded by Rick Rudman and Bob Lentz in
1992. The company, based in Lanham, Maryland, went public
in December 2005 and provides on-demand, Web-based
software for public relations. Their platform monitors news,
sends out press releases, and helps communicate with the media
and the public. Today they have over 2000 active customers
representing organizations of all sizes across a wide variety of
industries.

As a part of their solution, Vocus provides a proprietary
information database of over 800,000 journalists, analysts,
media outlets, and publicity opportunities. Their database
contains information about the media, including in-depth
journalist profiles, contact schedules, podcast interviews,
CLOUD



38
pitching preferences, and other relevant information compiled by
its dedicated media research team. The Vocus database is
integrated with a suite of on-demand modules that, together,
address the communications life cycle, from identifying key
contacts, to distributing information, to closing the loop with
digitized feedback and management analytics.

Vocus’s application is used by a wide variety of
organizations. After a successful career in television news at
ABC, Christine Nyirjesy Bragale, joined Goodwill Industries
International, one of the largest nonprofit organization in North
America, as international spokesperson and Director of Media
Relations. When she joined, all media relationships were
tracked and managed in a notebook (no, not a computer—a real
notebook). By replacing the notebook with Vocus they are able
to not only access records of interactions with journalist but also
be able to measure the results of news coverage. For a business
that is over $2B in revenue that’s no small step.

Omniture Story
Josh James and John Pestana met as undergraduates at a
computer science class at Brigham Young University. They
focused their first business on Web design, developer’s tools,
and gathering Web visitor data primarily for small businesses.
APPLICATION CLOUD



39
While it was a good business the cancellation rate for these small
businesses was so high they had to constantly run to stay in
place. Luckily, some large enterprise customers saw the value
of their software and it didn’t take them long to shift their focus
to providing Web analytics solutions for those larger customers.
Reflecting the new business focus, the company was renamed
Omniture in 1998. Omniture is derived from the words
“omniscient” and “future.”

Josh recalls the early days when they were offering software
as a service, before people even knew what to call it. “In the
early days, we had to fight through numerous issues, like
security, redundancy, and availability. We decided the best
thing to do was to go after the larger companies and then use
them as references.”

Today, Omniture is used in a wide variety of applications.
Overstock.com was losing customers in the checkout process.
Using Omniture’s SiteCatalyst software, they refined their
checkout process from seven down to three Web pages, resulting
in a 70 percent increase in conversion rates. In another
application, Novell uses Omniture SiteCatalyst on their
corporate intranet to track the use of applications, static Web
pages, and dynamic Web pages. Based on this data, Novell can
CLOUD



40
better expose pages, remove them, or adjust resources so that
they don’t have to spend money on pages or applications that are
not being accessed.

More recently, Omniture acquired Visual Sciences, which
had themselves, acquired WebSideStory and now finds itself as
one of the gorillas in Web analytics. Josh loves the subscription
business model, which he likens to a freight train. Since your
competitors cannot see the bookings, only the revenue, it seems
as if the business is growing slowly at first. But, "just like a
freight train, by the time you hear it, it’s rocketing past you—
and there is no way to catch up!"

Constant Contact Story
Constant Contact, Inc. was incorporated as Roving Software
in 1995. The company had just seven employees and a handful
of clients when Gail Goodman took the reins in 1999. Goodman
raised more than $20M in her first year and launched the
company’s first product: a permission-based email tool. Gail’s
advice to would be entrepreneurs: "Get a lot of advice, but don't
feel like you're a slave to every piece of advice you get. I
networked like crazy. I had never raised venture capital, so I
talked to lot of people who had done that. Some of the advice I
got was on target. Some of it wasn't."
APPLICATION CLOUD



41

By 2004, the company changed their name to Constant
Contact and in October 2007 completed its initial public
offering. Today, the company has over 150,000 worldwide
clients that depend on its Web-based email marketing service to
communicate with their customers. Reflecting on her years as
CEO Goodman said, “My most rewarding business moment …
was reaching 100,000 customers and having a really fun party."
And "my scariest business moment ... happened in summer 2001
with less than one payroll in the bank, writing the shutdown
plan. It was just a very dark week or two."

Constant Contact’s customers span the range, beginning with
an unlikely category - churches. Bethany Church of the
Nazarene has an active, dedicated congregation the church
inspires through innovative and creative communications. Bill
Miller, their communications pastor made it his mission to
update the way they communicated their message to the
congregation. Since they started sending out their new monthly
newsletter, the church has experienced a tremendous increase in
the number of people signing up. Not only does their newsletter
reach 3,000 opt-in subscribers weekly but it also reaches people
in multiple countries. “It is an incredible feeling to know that
people in another country want to be a part of our extended
CLOUD



42
family,” said Bill. “Knowing that the news of our congregation
and the work we’re doing has a far greater reach through our
newsletter than it ever could have otherwise is encouraging and
has a positive effect on everyone involved with Bethany.”

CRM for Sales
In most companies the largest single expense is sales. As a
result much of the initial focus of traditional CRM companies,
like Siebel and the leading application cloud service company,
salesforce.com, has been building software to increase sales
productivity. CRM for Sales seeks to standardize the sales
process and make it repetitive. Whether it’s having visibility to
a lead by status, the lead conversion%, converted leads by month
or who is the tops sales rep this class of software has been
adopted by small, medium and large enterprises.

Salesforce.com Story
The largest and most visible player in the application cloud
service market has been Salesforce.com. Marc Benioff founded
Salesforce.com in 1999 to pursue a simple idea: deliver
enterprise applications on the Web. With that vision, and with a
team of three software developers, Marc founded
Salesforce.com. The company began, not in a garage, but in a
rented San Francisco apartment, next door to his home. It
received no venture capital investment, but was funded instead
APPLICATION CLOUD



43
by Marc and Larry Ellison. Marc gave his company a mission—
to bring about the end of traditional software. Everywhere he
went he wore a pin that had the word software crossed out. At a
now-famous keynote speech, Marc constantly changed from a
business suit to a Hawaiian shirt and debated himself on the
merits of the old and new ways to deliver software.

From those early days, Salesforce.com’s growth has been
meteoric and has symbolized the movement to software on
demand. “Our growth validates the belief in ‘The end of
software,’ as well as raises new questions for existing software
companies,” says Marc. The company made an Initial Public
Offering in 2004 and, by 2005, was generating over $150M in
revenue. In 2009, they were the highest valued of all of the
application cloud service companies. There will be a few stories
about salesforce.com throughout the book. For the insider’s view
of the creation of salesforce.com check out Marc’s new book
3
.

CRM for Service
Once you have a customer increasingly keeping a customer is
defined by the quality of the service delivered. We all know that
high quality service is relevant information, personal to you,

3
Benioff, M., Behind the Cloud: The Untold Story of How Salesforce.com Went from Idea to
Billion-Dollar Company-and Revolutionized an Industry, Jossey-Bass (October 19, 2009).

CLOUD



44
delivered, as you need it. Whether thru web, chat, email or
phone service is all about information.

At the highest level, the U.S. economy is a service economy.
Figure 2.1 shows the dramatic shift from manufacturing and
agriculture to services over the last 50 years.


Figure 2.1 Growth of the U.S. Services Economy

Today’s services economy accounts for 80 percent of the U.S.
GDP (gross domestic product). Financial services, healthcare,
education, IT services, and software are all industries that don’t
manufacture or grow a single thing. Furthermore, contrary to
conventional wisdom, this shift has not been an increase in the
number of kids flipping hamburgers. Instead, 70 percent of the
service employment involves highly skilled people. This
APPLICATION CLOUD



45
includes executives, scientists, and administration and
management functions. Figure 2.2 shows the results of the 1996
census.


Figure 2.2 United States Census (1996)

Services economy jobs include software designers, hedge
fund managers, doctors, lawyers, and customer service
professionals and all of these jobs are fundamentally different
than working on the assembly line in the early 20th century.

One of the earliest companies to focus on customer service
application was RightNow Technologies.

RightNow Technologies Story
Founded by Greg Gianforte in 1997, RightNow went public
in August 2004. Based in Bozeman, Montana, RightNow
focused on helping customer service organizations improve
service delivery using technology. Gianforte didn’t start
CLOUD



46
RightNow with venture capital, and makes a passionate case for
starting businesses the old fashioned way. He published a book
4

in which he builds a strong case against relying on venture
capital (VC) funding:
• Raising money takes time away from understanding your
market and potential customers. Often more time than it
would take to just go sell something to a customer. Let
your customers fund your business through product
orders.
• Money removes spending discipline. If you have the
money you will spend it—whether you have figured out
your business model and market or not.
• You sell your precious equity very dearly before you have
a proven business model. This is the worst time to raise
money from a valuation perspective.

Gianforte knows he is pushing a contrarian view, but he
reminds the reader that Dell, HP, and Microsoft all originally
started without VC funding. In RightNow’s case, they doubled
their revenues and employees every 90 days for two years before
they took any outside money, and, even then, the employees
retained more than 75 percent ownership after the company
raised $32M.


4
Gianforte, G. "Bootstrapping Your Business: Start and Grow a Successful Company with
Almost No Money.” BookSurge Publishing, 2007
APPLICATION CLOUD



47
RightNow provides a suite of customer service and support
solutions that captures, tracks, assigns, and manages customer
service interactions from initial contact through resolution.
Using their case management, Web self-service, email
management, and collaboration modules, companies can more
effectively deliver support and service for their customers.
Some of RightNow’s customers include Ken Harris, CIO at
Shaklee, a 50 year old natural nutrition company; Boyd Beasly,
Senior Director of Customer Support at EA, a $3B+ leader in
electronic gaming and Chuck Udzinski, Consumer Services
Manager at Black & Decker, a $6B+ manufacturer of power
tools.

“Delivering applications as a cloud service is more than just
another way to implement software. It fundamentally changes
both the relationship between the software vendor and the
customer, and the economics of purchasing and owning
software,” says Gianforte. “Customers want solutions that are
up and running in days or weeks, not months or years, and they
are looking for ways to reduce overall ownership costs.
Utilizing cloud delivery models, organizations are able to get
these advantages over on-premise applications. The cloud is
altering the way enterprise applications are being consumed, and
CLOUD



48
today we are just in the early stages of this next evolutionary
stage for software.”

Vertical Applications
Reading Jared Diamond’s bestseller, “Guns, Germs and Steel:
The Fates of Human Societies,”
5
may make you see the world a
little differently. Diamond’s book centers around answering a
question he was asked by his New Guinea friend Yali: “Why is
it that you white people developed so much cargo [goods] and
brought it to New Guinea, but we black people had little cargo of
our own?” The answer is that, due to conditions in a few
isolated places in the ancient world, some peoples began to
produce food either through herding large mammals or growing
crops. The lucky ones had the right circumstances to do both.
Being a food producer meant that the population could grow.
Having the ability to store food meant that the society could
develop specialization.

Specialization meant technology could develop. Some people
could make weapons, others could grow food, and still others
could erect buildings. This is no surprise to us in the modern
world of work, but what does specialization mean in software,
and why does specialization matter? Probably the best example

5 Diamond, Jared, “Guns, Germs, and Steel: The Fates of Human Societies”, W.W. Norton &
Co.; 1 edition (July 11, 2005)
APPLICATION CLOUD



49
of the power of specialization is Google. No one had to teach
anyone how to use the system; it became immediately obvious
what it did and, to the teams building the software, what they
needed to do. If Google had started out to build software that
enabled you to buy a book, auction a Pez dispenser and search,
we’re sure history would be quite different. So how specialized
can software get? We’re conditioned to think that business
software is about ERP and CRM, that if we provide purchasing,
HR, financial, and sales force automation solutions, we’ve built
all the software a business needs. But if you walk into any
ongoing business, you’ll discover that the ERP/CRM footprint is
less than ten percent of their overall software. Consider the fact
that CitiGroup today employs more programmers than SAP or
Oracle.

A slightly tongue in cheek example of specialized software is
Flock IT.

CLOUD



50

Figure 2.3 Specializations

Flock IT is a company specialized in poultry management
software. In the not so distant past, the cost of supporting and
selling this software (pre-Internet) would have been so high that
the founder could only support customers in his hometown of
Brisbane, Australia. Had he wanted to expand to Sydney, he
would have had to hire a support engineer and a salesperson.
Unless he had enough capital and a high enough price on the
software, he just couldn’t afford to do this. Search Yahoo! or
Google and you’ll find thousands and thousands on Flock IT-
like companies.

But post-Internet, do we have to think this way? On a global
basis, are there not enough people managing poultry to use this
software? And once you see this possibility, doesn’t the future
APPLICATION CLOUD



51
look like lots and lots of specialized software: search,
auctioning, cattle management, pig management, and poultry
management software?

There are some many examples of specialized applications
including those provided by Blackbaud for nonprofit
organizations, Dealertrack for automotive dealerships and
Opentable for reservation taking restaurants.

DealerTrack Story
DealerTrack was founded in January 2001 with the initial
financing of $20M coming from three forward thinking financial
institutions: J.P. Morgan Chase & Co., AmeriCredit Corp., and
Wells Fargo & Company. They saw whenever a consumer
applied for an auto loan; the old system was both slow and
manually driven, with each lending institution using a different
form—cumbersome for both the consumer and the dealer. In
their first year, DealerTrack brought on line over 20 percent of
the 22,000 dealers in the United States, but by late 2002, a dark
cloud came onto the horizon. Chrysler, Ford and General
Motors decided to finance a new company, RouteOne, to the
tune of $100M. If anything created focus at DealerTrack, it was
this much-needed competition. The company began to add more
dealers (60 percent signed up in year three) as well as more
CLOUD



52
products for the dealers. DealerTrack went public in 2005, and
today the company is connected to almost all of the 22,000
dealers in the U.S. More interesting, DealerTrack has
established an electronic channel, a real “information
superhighway,” that allows them to continue to distribute new
software products, which they are doing at the rate of three per
year.

When Mark O’Neil, Chairman and CEO of DealerTrack,
talks about delivering software as a service, he will say there is
no other way to run a software business. Not only does he get
tremendous economic/cost leverage, which gives him
tremendous pricing leverage; but the application cloud service
model also gives the company the ability to release new products
quickly and nimbly.

Blackbaud Story
Founded by a soccer-playing Englishman, Blackbaud is the
market leader in software for non-profit organizations. Their
story contains elements of all of the seven business models.
First, they have a large support and maintenance revenue stream
as a by-product of selling in traditional Model One. In addition
a few years ago they began offering their traditional products in
Model Four. Below is a chart of the growth of that business.
APPLICATION CLOUD



53


Figure 2.4 Blackbaud’s Four Product Lines

Finally, thru acquisitions in 2006 of eTapestry and 2008 of
Kintera, Blackbaud has two Model Six companies engineered
for multi-tenant low cost delivery.

OpenTable Story
As one of the newest public company Opentable is a great
example of the blurring of the boundaries between consumer and
enterprise software. On the consumer side, OpenTable allows
you to find and make reservations at restaurants based on time,
date, cuisine, and price range. Upon completing the reservation
you receive OpenTable rewards points, which can be redeemed
for discounts at member restaurants. On the enterprise side
Opentable provides restaurant owners with a reservation
management system designed to replace existing paper
reservation systems. The solution is an integrated software and
CLOUD



54
hardware solution that computerizes restaurant host-stand
operations. The system keeps track of VIPs, customer
preferences, integrates with the POS system and allows for re-
configuring the tables for larger and smaller seating
arrangements. Given the relative unreliability of the network
connections to the implementation is a standardized PC.

In many ways the business looks like a combination of a
Model One and a Model Seven business. Sales are done using
the traditional door-to-door approach and once the buyer says
yes, a consulting arrangement occurs to implement the
reservation system. On the other hand the consumer side of the
business is a classic Model Seven business. As a diner, you don’t
pay for the service. Instead Opentable charges the restaurant a
monthly fee to be part of the Opentable network as well as a
transaction fee that varies depending on whether the reservation
originated from Opentable’s site or the restaurant’s site.

Cloud Service Stack
Many of these application software companies were
architected in the late 1990s. But if you were going to start today
there is a much larger array of services that you might consider
using. We’ve developed a five-layer stack of services beginning
with platform cloud services and including compute & storage
APPLICATION CLOUD



55
cloud services, co-location services and the original network
cloud services.

Network Cloud Services
Co-Location Cloud Services
Compute & Storage Cloud Services
Platform Cloud Services
Application Cloud Services
Network Cloud Services
Co-Location Cloud Services
Compute & Storage Cloud Services
Platform Cloud Services
Application Cloud Services

Figure 2.5 Cloud Service Stack

Platform cloud services are either service used by software
developers to build new applications (e.g., Microsoft Azure,
Google AppEngine or Netsuite’s NSBOS), or operations staff to
manage these application cloud services (e.g., Postini for spam
filtering or Service.now for trouble ticketing). Amazon Web
Services pioneered compute & storage cloud services, which
give you access to computer and storage resources. Compute &
storage services use co-location services to house, power and
secure the physical hardware. Finally all of these services are
connected thru network cloud services. This organization will
serve to frame the next several chapters.

CLOUD



56
Summary
The early pioneers of SaaS/Application cloud services have
demonstrated there is the ability to sell and deliver applications
in the new business models. These early pioneers had to build
much of the infrastructure and software to deliver and sell these
applications. What’s exciting today is the creation of 1000s of
new cloud services which new providers of applications can now
take advantage of. So whether you’re taking an existing
application and delivering it as a cloud service, developing one
for in-house use, or trying to be the next public company the
good news is these new cloud services can speed the delivery
and lower the cost of getting new applications into the market.
The next several chapters will focus on the different layers of
this new cloud stack.











57

DATA CENTER


The compute, storage and network hardware that power all
cloud services ultimately must reside in a facility, connected to
the network. If you choose to build your own compute and
storage cloud service; or if you consume application or platform
cloud services this chapter is written to highlight the major areas
you’ll need to understand.

We'll begin where the name “cloud” began, which are the
basic network connection services. We’ll move on to discussing
one of the key attributes in the reliability and cost of any cloud
service: power. Staying with the physical infrastructure, we’ll
CLOUD



58
highlight some of the key attributes in building highly available
and secure facilities and conclude with the fundamental
economics of building data centers. At $1,000 per square foot
it’s some of the most expensive commercial real estate you’ll
ever see.

Network Cloud Services
Many have forgotten, but the reason we all call this cloud
computing is because we always drew the network (and the
Internet) as a cloud. By that we meant, we weren’t sure how it
worked but somehow bits were transmitted from one point to
another, reliably and efficiently.

Clearly, entire books can be written on network cloud service
so for our purposes we’re going to focus on one valuable idea.
Anyone operating an application cloud service will one day need
to operate globally, and will require at least one disaster
recovery site. If networks were indeed uniformly cheap, reliable
and fast then operating out of 2 centers on the planet could work.
However, this is not the case so it will be important that the
application deployments are location independent. It is here that
the world of appliances and application cloud services intersect.

LOCATION, LOCATION, LOCATION



59
At some point in the day or night, the people managing the
application will be remote from the physical location of the
computer. If you want the people managing the software to be
working the first shift, versus keeping a team in the US on three
shifts, then you’ll locate people in the Americas, Europe and
Asia time zones. Once you’ve done that, you will need to
establish a “Back of the Box” Network to connect the computers
to the operations staff, as shown in Figure 3.1.


Figure 3.1 Front of the Box, Back of the Box

Your next question will be where to locate the computers. If
you want to provide for a disaster recovery site, you’ll need at
least two sites. Beyond that, you’ll need to figure out where your
users are concentrated, and the cost, performance and reliability
of your “Front of the Box” Network. Companies like Akamai
CLOUD



60
have built a large business helping companies provide more
reliable, faster connections. But, in the end, if you’re application
needs to run in Chile, the cost of bandwidth might be such that
you’ll need to locate the computers in Chile. Again, this should
be no problem if you have architected a location-independent
application.

Furthermore, as soon as you realize that standardizing the
infrastructure of software and hardware to run in Shanghai, San
Francisco and Paris is critical to global deployment; it really
doesn’t matter where the building that houses the computers is
located. Furthermore, it becomes only a financing question as to
who owns the actual data center. As a result, there is a fine line
between application appliances, and a multi-location deployment
of any application cloud service. It becomes more a matter of
who pays for what, as opposed to the operational technology.
There are, of course, some downsides to not being able to keep
all of the hardware in a few locations. In particular, it will be
more expensive, as it will be much more difficult to take
advantage of spare capacity afforded by other users consuming
the same compute and storage resources.
LOCATION, LOCATION, LOCATION



61


NTT Story
While adoption of business applications delivered out of the
cloud has lagged in Europe, NTT Europe Online is trying to
change all of that. NTT Europe Online is a wholly owned
subsidiary of Nippon Telephone and Telegraph, the largest
telecommunications company in the world. They provide
private cloud services from five data centers across Europe: UK,
France, Germany, Spain, and Switzerland. Today their
customers’ ranges from Web sites such as Nissan Europe to
specialized software companies like Aspiren, and EnergySys.

Aspiren, which focuses on business performance
management software for local and central governments,
contracted with the UK government to provide the National
Performance Management Framework (NPMF). As part of the
contract they are delivering a service for 408 local government
authorities to benchmark and improve performance, share
leading practices, deliver efficiency savings, improve service
levels, and report electronically to the central government. You
can only imagine the cost and complexity of rolling out software
to hundreds local government offices in the traditional Model
CLOUD



62
One world—or should we say—the help desk phone would
probably have melted.

EnergySys provides carbon-reporting and accounting
software for the oil and gas industry. One of their customers is
the South Caucasus Pipeline Company Limited, which transports
gas to markets in Azerbaijan, Georgia, and Turkey. Using the
EnergySys solution they were able to go live in months.
EnergySys gives their customers the choice of having the
application delivered as a cloud service or as an appliance the
customers can host themselves. But even if the customer
chooses the appliance to be installed “@Customer” EnergSys
manages the software and insures changes, performance and
compatibility are managed. They also price the appliance the
same as the online system—the only difference is a minimum
that reflects the cost of a dedicated solution. Peter Black, CEO
of EnergySys, who started out the business delivering software
in Model One, now says—“I don’t know why all software is not
delivered as a service”. And with NTT Europe Online’s help
perhaps much more software developed and deployed in Europe
will follow Peter’s lead.

LOCATION, LOCATION, LOCATION



63
Power Quality
Computers and networks cannot operate without power, and
no matter how reliable your electric grid, there is always the
possibility you will lose power. During the past decade, there
have been several efforts to assess the national cost of power
interruptions and power quality. The Electric Power Research
Institute (EPRI) estimates the cost at over $119B per year.
6

EPRI, and the National Power Lab data also shows that a typical
facility will on average have a voltage fluctuation lasting
between one cycle to over thirty minutes 38 times per year. In
fact at least 4 times per year the power will be out for a minute
or more.

As a result, quality data centers will be equipped with UPS
and diesel generators. You should, of course periodically run a
test to see if the generators are working—and make sure you run
the test long enough. A few years ago, a large data center had to
rely on their diesel generators. After two minutes, the systems
failed, even though they had tested the generators every quarter.
It turned out their quarterly test duration was for 30 seconds.
Unfortunately there was a clog in the fuel line, but during testing
there was enough fuel in the line to run the test for 30 seconds,
and no more.

6
The Cost of Power Disturbances to Industrial and Digital Economy Companies. Madison WI.
(TR-1006274, Available through EPRI). June 29, 2001,
CLOUD



64

Power Utilization Efficiency
While having high quality power is important, as we will see
the effective usage of power has both economic and carbon
impacts. The Power Usage Effectiveness
7
(PUE) metric is
defined as the ratio of the total power consumed by a data center
to the power consumed by the IT equipment that populates the
facility. PUE was established by The Green Grid, an association
of IT professionals seeking to dramatically raise the energy
efficiency of data centers through a series of short-term and
long-term proposals. To give you a sense of the range of
possibilities, the Uptime Institute gathered data from many of
the 85 members of the Institute’s Site Uptime Network®. In
their sample size the average PUE was 2.58.

In another published work
9
a team from the Lawrence
Berkeley National Labs benchmarked 22 data centers. Their
analysis showed data centers could be over 40 times as energy
intensive as conventional office buildings. They created a
metric defined as the Computer Power Consumption Index,

7
The Green Grid, 2007, “The Green Grid Data Center Power Efficiency Metrics: PUE and
DCiE,” Technical Committee White Paper.
8
Data Center Energy Efficiency and Productivity By Kenneth G. Brill , Presented at The
Uptime Institute Symposium 2007
9
Best Practices for Data Centers: Lessons Learned from Benchmarking 22 Data Centers Steve
Greenberg, Evan Mills, and Bill Tschudi, Lawrence Berkeley National Laboratory Peter Rumsey,
Rumsey Engineers Bruce Myatt, EYP Mission Critical Facilities
LOCATION, LOCATION, LOCATION



65
which as it turns out is the inverse of the PUE. Figure 3.2 shows
the Computer Power Consumption Index for all 22 data centers
in their study. The 22 data centers ranged from a PUE of 1.5 to
3.0.

Figure 3.2: Computer Power Consumption Index for 22 Data Centers
(Inverse of PUE)

Again, because large Model 7 companies consume large
numbers of servers per employee it makes sense that companies
like Google are pushing the state of the art. Figure 3.3
summarizes the PUE from all Google-designed data centers with
an IT load of at least 5MW and time-in-operation of at least 6
months.

CLOUD



66

Figure 3.3: Google Data Center PUE

Why is there variation among the data? Power and cooling
architectures differ between facilities, and the facilities
themselves are located in different a climate, which influences
the PUE performance. Also, the data centers shown were built at
various times since 2005, and over time their designs have
become more efficient. For example, Data Center A is the oldest
facility of the group and has one of the highest average PUE
values. Data Centers E and F are two of the newest and have
among the lowest values. In addition, PUE values are impacted
by seasonal weather patterns, and thus the PUE during cooler
quarters tends to be lower than in warmer ones.

LOCATION, LOCATION, LOCATION



67
Power Carbon Footprint
With the increasing cost of power and the need to reduce
carbon emissions, the other major area of focus is the reduction
of overall power and cooling requirements. Dr. Jonathan G.
Koomey completed a study on the global power consumption of
servers. The total power used by servers represented about 0.6
percent of total U.S. electricity consumption in 2005. When
cooling and auxiliary infrastructure are included, that number
grows to 1.2 percent, an amount comparable to color televisions’
power consumption. In total power demand, this is equivalent to
about five 1000-MW power plants for the U.S. and 14 such
plants for the entire world. The total electricity bill for operating
those servers and associated infrastructure in 2005 was about
$2.7B and $7.2B for the U.S. and the world, respectively.
10


Not all power is created equal. Some power has a much
higher carbon footprint than other. For this answer we turn to the
Emissions & Generation Resource Integrated Database
11

(eGRID). eGRID is a comprehensive source of data on the
environmental characteristics of almost all electric power
generated in the United States.


10
Koomey, Jonathan. Estimating Total Power Consumption by Servers in the U.S. and the
World. Stanford University, February 15, 2007.
11
http://cfpub.epa.gov/egridweb/
CLOUD



68
The annual total output emission rate is the measure of the
carbon emissions as it relates to the generation output. Units are
in lb/MWh and are converted to lb/Watt-year by multiplying by
0.00876. If you go thru the database you will find specific
power plants that range from 0 to 876-lbs/watt year. Surprisingly
there are coal-fired plants that are cleaner than some oil-fired
plans. If you know which specific power plant your data center
is connected to then that CPF can be used, otherwise eGRID
divides the country into a number of sub-regions.

Figure 3.4: Sub-region power grids

Interestingly the sub-regions have a nearly 4:1 difference in
the amount of carbon generated per Watt (40-170 pounds/Watt-
year).

LOCATION, LOCATION, LOCATION



69
PG&E Story
Pacific Gas & Electric (PG&E), one of the largest utilities in
the United States, is offering significant rebates as part of a new
energy incentive program for computer servers. As part of
PG&E’s Non Residential Retrofit program, customers replacing
existing equipment with new servers from Sun Microsystems
can receive a cash savings of between $700 and $1000 per
server. This is the first-ever incentive rebate offered by a public
utility company for servers. Perhaps we’ll see more of this in
the future.

Availability Management
Whether you decide you need to own your own buildings or
just rent, obviously one of the important criteria for a co-location
cloud service is the availability. The Uptime Institute has
identified four levels of availability – the best being Tier 4, the
lowest level being Tier 1.

Tier 1 data centers have an average outage profile of 1700
outage minutes per year. Tier 1 co-location cloud services may
or may not have UPS, or generators and typically will have only
single paths for power and cooling. Tier 2 data centers provide
1300 outage minutes per year and again will typically require
maintenance of power to require a shutdown. Tier 3 co-location
CLOUD



70
services move the outage performance significantly and are
designed to deliver 90 outage minutes per year. In order to be
able to do this there will be multiple power and cooling
distribution paths. Finally, Tier 4 facilities are the top of the heap
delivering 26 outage minutes per year.

Security Management
The security of a co-location cloud service is has both a
physical and a people component. In general the highly secure
facilities will have few windows, a 100-foot vehicle buffer zone,
foot thick concrete walls and retractable crash barriers. There
will be few entry points and for data centers that are especially
sensitive or likely targets, guards may use mirrors to check
underneath vehicles for explosives. Surveillance cameras will be
installed around the perimeter of the building, at all entrances
and exits, and at every access point throughout the building.

This brings us to the people side of the equation. All
personnel working in the facility should have background
checks. Furthermore using two-factor or biometric identification
is becoming standard for access to sensitive areas of data
centers. Finally for authenticated individuals, limiting access to
areas on a need to access basis.

LOCATION, LOCATION, LOCATION



71
AT&T Story
In many ways, USi was a creation, and later a victim, of the
Internet bubble. Stock in the once publicly held company,
reached $66 a share in early 2000. Fleets of jet skis were docked
nearby for employees eager for a spin on the Severn River.
Payroll ballooned to 1,400 employees. USi raced to build
expensive data centers to support the wide range of software it
offered.

When the bubble burst, USi was hit hard, ultimately filing for
bankruptcy in January 2002. AT&T acquired USi in October
2006 for $300M. Today AT&T’s services over 150 clients with
end users in over 30 countries. HealthMedia, Inc. is a good
example of a software company that used AT&T’s co-location
services. HealthMedia's software application provides online
programs that focus on disease management (e.g., diabetes), pain
management (e.g., arthritis, lower back pain), weight
management, and other wellness programs. Companies like
Jenny Craig, GlaxoSmithKline, Pfizer, and UPS provide some of
these programs to their employees. Since they are in the health
care industry, the company must comply with several unique
requirements, including the Health Insurance Portability and
Accountability Act (HIPAA) and Title 21 Code of Federal
Regulations. As a small company there was no way compliance
CLOUD



72
was possible, but leveraging the scale and expertise of AT&T
they could focus on their software and leave the data center
certification to their cloud provider.

Economics
Building data centers is expensive. In 2007 Microsoft decided
to build a 470,000 sq foot data center at an estimated cost of
$550M. In 2008 Amazon.com broke ground on an 116,000 sq
foot data center at a cost of $100M. And just so you don’t think
this is purely a US phenomenon, GDS-China is in the process of
building five new data centers in China, with a plan for fifteen
more. At ~$1,000 per square foot these are expensive capital
investments. In Figure 3.5 you’ll see an IBM estimate of the cost
breakdown for the various components. Sixty percent of the cost
is mechanical and power systems.

Figure 3.5 Co-location Facility Capital

LOCATION, LOCATION, LOCATION



73
This brings us to the on-going operational costs of a data
center. Figure 3.6 is an analysis done by IBM which shows what
many who have been students of this area already know. That is,
that power or energy costs dominate the operational costs of a
data center.
$50
$100
$150
$200
$250
$300
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
C
u
m
u
l
a
t
i
v
e

C
o
s
t

o
f

O
p
e
r
a
t
i
o
n
s

(
$
m
)
Year
Source: IBM engineering estimates, 2008
Energy Cost
Staffing
Bldg. Maint. & Mgmt.
R E Tax
$50
$100
$150
$200
$250
$300
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
C
u
m
u
l
a
t
i
v
e

C
o
s
t

o
f

O
p
e
r
a
t
i
o
n
s

(
$
m
)
Year
Source: IBM engineering estimates, 2008
Energy Cost
Staffing
Bldg. Maint. & Mgmt.
R E Tax
Energy Cost
Staffing
Bldg. Maint. & Mgmt.
R E Tax
Energy Cost
Staffing
Bldg. Maint. & Mgmt.
R E Tax

Figure 3.6 Co-location Operational Costs

Christian Beladay’s analysis
12
shows that as of 2008 energy
costs alone have exceeded server costs. Moreover the cost of
power can fluctuate by a factor of at least 6x, from 3 cents per
KWh in Idaho to 18 cents in Hawaii. So in the end it may turn
out that future data centers will just be value add to anyone
generating power.


12
Belady, C. “In the Data Center, Power and Cooling Costs More than IT equipment it supports”,
Electronics Cooling Magazine.
CLOUD



74
Summary
Network and co-location cloud services represent significant
capital investments to build out. Increasingly the location of the
data centers will be dictated not only by the proximity to
networks, but also by the availability of low-cost, low-carbon
high quality power. These data centers will in many ways be the
bridges and roads of the next generation of all industries.
Utilizing these co-location cloud services the next layer of the
stack is to provide compute and storage cloud services, i.e.,
providing computers and storage technology purchased by the
hour, or the month. We’ll discuss this more on the next chapter.









75

COMPUTE & STORAGE CLOUD
SERVICES


The cover story in Business Week in November 2006 was
entitled Jeff Bezos’ Risky Bet
13
. The article focused on Bezos’
vision of transforming the Internet retail giant into a cloud
service provider. It started with the Amazon eCommerce API
and Alexa Web Search services. But not until the Simple Storage
Service (S3) rolled out, did it start to become a reality and with
the creation of Elastic Compute Cloud (EC2), Amazon began
another revolution.


13
Jeff Bezos’ Risky Bet, Business Week November 2006
CLOUD



76
Compute & Storage Cloud Services
Amazon’s Elastic Compute Cloud (EC2) and Simple Storage
Service (S3) were the first compute and storage cloud services.
Compute cloud services allows scalable deployment of
applications by providing a web service through which a user
create a virtual machine instance containing a wide-range of
software. Typically these virtual machines include pre-
configured Solaris, Windows, or Linux images. While
pioneered by Amazon, compute cloud services are now available
from Microsoft, Rackspace and Opsource.

A compute cloud service enables people to create, launch,
and terminate server instances as needed, paying by the hour,
starting at around $0.10. Some services also provide control over
the geographical location of instances, which allows for latency
optimization and high levels of redundancy. Also these new
compute cloud services will allow you to choose the size of
memory as well as the number of virtual cores, all for an
increase in the hourly fee.

Along with compute services are storage services. In fact
Amazon’s S3 pre-dates the availability of EC2. Storage cloud
services provide "unlimited online storage" again in a
subscription model (typically $0.15/GB/month). Very typically
COMPUTE & STORAGE


77
the storage services keep data in at least two, if not three
different data centers. The cost of online storage is dramatically
lower than tape; hence this architecture precludes the need to
provide tape backup and offsite storage.

Amazon Story
At the 2007 Web 2.0 conference, Jeff Bezos delivered the
keynote speech and shared some pretty astounding numbers for
S3, including a peak day when there were 921 million requests.
The number of Amazon Web Service users has grown from
240,000 in 2007 to 540,000 in Q1 of 2009, and in the same time
period the number of storage objects has gone from five billion
in 2007 to over fifty billion.

In keeping with their growing sophistication in 2008, EC2
provided a service level agreement, a formal commitment that
the service will be available at least 99.95 percent of the time.
With growing usage also comes growing sophistication in the
management infrastructure.

Chief Technology Officer Werner Vogel described some of
Amazon's continued investment in reliability and manageability:
“We relentlessly measure every possible resource usage
parameter, every application counter, and every customer's
experience. Many gigabits per second of monitoring data flow
CLOUD



78
continuously through the Amazon networks to make sure that
our customers are getting serviced at the levels they can expect
and at efficiency level the business desires.”

Beyond S3 and EC2 Amazon introduced other services
including CloudFront, SimpleDB, and Public Data Sets.
CloudFront caches high-traffic content on the company's
worldwide network of edge locations so it's always near the end-
user for low latency. Customers store the original versions of
content (photos, video, music, applications, etc.) into S3, call up
an API that returns a unique domain name for the content, then
link to it on a Web site or Web application. Amazon has 14
edge locations around the world including Dallas, Los Angeles,
Miami, Newark, Palo Alto, Seattle, St. Louis, Amsterdam,
Dublin, Frankfurt, London, Hong Kong, and Tokyo. SimpleDB
is a new service providing simple relational database service.
SimpleDB general manager, Matt Domo says, “Amazon
SimpleDB eliminates complexity so that developers and
businesses can focus on optimizing applications and not
administering their database.”

In addition to their own efforts, Amazon is creating a partner
community that is further extending their cloud capability.
Vertica Systems offers their highly scalable, columnar DBMS on
COMPUTE & STORAGE


79
EC2. It enables customers to automatically configure in a matter
of minutes analytic databases of virtually unlimited size and
performance at a fraction of the cost of conventional systems.
Jerry Held, Chairman of Vertica, has been a major figure in the
evolution of database technology. As a Ph.D. student at Cal he
worked on the relational database that would come to be known
as Ingres. He was one of the first developers at Tandem
Computers and wrote the original file system and ultimately he
went on to preside over Oracle’s billion-dollar database
business. With that breadth of experience he joined Vertica as
Chairman of the Board and has been instrumental in moving
them towards the cloud. Jerry says, “The combination of
Vertica’s ultra high performance at extremely low price and
EC2’s dynamic configurability are enabling business changing
analytics that haven’t previously been conceivable. Bringing
analytic applications on-line in hours rather than weeks or
months and having access to huge resources for varying
durations means that what would have been a multi-million
dollar capital investment can be turned into a short term, small,
opex decision which can be made by line management.” And he
should know.

But these cloud services are not only being used by new
startup companies but also by Fortune 500 corporations
including Eli Lilly and Autodesk. “This is a huge step forward
CLOUD



80
in maximizing our results relative to IT spend, and now that
Amazon EC2 runs Windows and SQL Server, we have even
greater flexibility in the kinds of applications we can build in the
AWS cloud,” said Dave Powers, an Eli Lilly associate
information consultant who uses the service to process research
data. “Autodesk uses EC2 for back-end data processing tasks,”
said Mike Haley, a senior architect of search engineering.

Further signaling the progress they are making was the recent
announcement from Capgemini. Capgemini will offer business
users Microsoft Sharepoint, Oracle ERP and application
development and testing in the cloud. In principle Capgemini is
running a Model Three business but using Amazon’s compute
and storage cloud to offer its outsourcing clients a lower cost
approach to pilot projects and IT projects. According to Andrew
Gough, SaaS business development manager at Capgemini.
“Every business has a project that never surfaces because 80%
of an IT budget goes on maintenance,” Gough said.

Virtual Machine
Compute cloud services are provided as virtual machine
instances. A virtual machine allows the sharing of the underlying
physical machine resources between different virtual machines,
each running its own operating system. The software layer
COMPUTE & STORAGE


81
providing the virtualization is called a virtual machine monitor
or hypervisor. VMWare is the leading company in providing
this type of software, although many other companies including
Microsoft and Oracle are providing their own. Amazon uses a
variation of the Zen open source implementation to deliver their
cloud services.

Virtual machines are very useful in situations where the
server utilization is low as it allows fewer servers to be shared,
thereby reducing power, cooling and space requirements. Figure
4.1 shows the server utilization across a typical data center.


Figure 4.1 Server Utilization in Data Centers

CLOUD



82
As a result, even in non-cloud implementations virtual
machines are being used for server consolidation as a means of
reducing cost, and in some cases carbon impact.

Security Management
As more compute & storage service providers enter the
market, some will differentiate purely on price, others on quality
of service and still others by geographic or industry
specializations. One area of particular customer concern is
security management. When providing security management at
the compute and storage level one needs the ability to robustly
authenticate anyone with logical or physical access; provide
fine-grained access control and finally audit the interactions. As
a result some vendors are providing dedicated VLANs, sub-
administrators (so there can be one master account, but multiple
groups using it), role-based user permissions and the ability to
track actions by department. Furthermore since one of the
major principles of good security management is auditability,
standards like SAS 70 Type I and Type II, HIPAA (for
healthcare) and PCI (for payment) compliance further
differentiate one offering from another.
.
COMPUTE & STORAGE


83
TD Banknorth Story
As a financial institution, TD Banknorth’s evaluation of
vendors’ offerings had to be more thorough than most.
Regulations imposed by the Office of the Comptroller of the
Currency, for example, set forth strict guidelines about what
constitutes appropriate due diligence in such a selection process.
In the end, they chose the RightNow service, and particularly
important for TD Banknorth was the SAS 70 report. Mark Ellis,
Senior Vice President of eCommerce at TD Banknorth, said they
“especially liked the fact that RightNow uses AT&T’s facilities,
which are subject to regular SAS 70 Level II audits.”

This, perhaps, makes a more important point. Anyone
delivering services in Models Four through Model Seven can
make investments in security on behalf of many customers and,
as a result, can arguably provide even higher levels of service
than any one customer could on their own.

Public and Private Clouds
Being able to control security, specification and performance
of the compute and storage infrastructure has given rise to the
terminology of public vs. private clouds. Some would say that
private clouds are just a marketing term to rename a company’s
existing data center. While this may be true it’s better again to
see the range of possibilities as a continuum rather than a black
CLOUD



84
and white decision. Public clouds, like public swimming pools,
are specified by the service provider. Of course within those
specifications the providers continue to provide increasing
degrees of flexibility and security, e.g., greater degrees of
control on who has access to what. Early implementations of the
public cloud didn’t allow you to insure there was capacity when
you needed it; however, more recently the idea of reserved
instances has been implemented where the user can lock away
compute capacity for a year or even three years.

However, if the specification of the cloud service does
not fit your needs then, just like with swimming pools, it’s
clearly possible to implement private clouds. As an example
consider JDA Software’s market for implementation of its
AS/400 E3 product. With nearly 1000 customers worldwide, an
implementation of a highly standardized AS/400 compute and
storage private cloud would clearly reduce the cost and improve
the reliability for any of their customers.
.
IBM Story
In February 2009, IBM’s CEO Sam Palmisano appointed
Erich Clementi, to head up cloud computing. IBM rarely makes
moves like this for strictly marketing purposes and in many
ways it signals IBM’s view that cloud computing represents the
COMPUTE & STORAGE


85
third major wave of business computing – preceded by client-
server and mainframe computing.

But why? In 2009 Erich Clementi presented IBM’s analysis
of the cloud opportunity. In short, their models predict the
market is growing at 28% (2008-2012 CAGR), compared to the
traditional IT market, which is growing very slowly (1% for
traditional SW and SVCS) or decreasing (-3% for traditional
HW). In 2012, they predict the cloud opportunity is substantial
($126B), and composed by three sub-markets:

1. Components: $44B. Includes all hardware, software
and services purchased by enterprises and direct cloud
providers to provide application, platform, compute &
storage cloud services to internal or external
customers. In effect this is being in the arms dealer
business selling traditional hardware and software.
2. Infrastructure services: $30B. Includes the revenue
derived by providers of platform, compute & storage
cloud services. Within compute & storage cloud
services a large portion of business will be driven by
clients not buying new physical infrastructure,
especially commodity x86 servers and storage, but
using public or private services. Here the shift from
traditional to new will appear to be even more
dramatic, as especially for public cloud services the
buying behavior shifts from one time investments to
small portions of recurring (or even non-recurring)
revenue, e.g. multiples of 10 cents per hour acquired
by many individuals unpredictable over the year.
3. Business services: $52B. Includes all applications and
any business process outsourcing built on top of
CLOUD



86
application cloud services. As an example IBM uses
Taleo to provide certain HR BPO services.

In the mid- and long-term solutions for various industry
verticals, small businesses and in the growth markets will benefit
dramatically from the new economics and the user experience of
cloud. As a result IBM is developing and delivering cloud
services across the spectrum, from moving their own global
development labs to using a compute & storage cloud service, to
providing a new horizontal platform service like Pangoo,
developed in IBM’s China Research Lab or delivering
application cloud services like LotusLive is a collaboration tool
that lets people share contacts and documents, hold online
meetings, and build social-networking communities. And even
beyond LotusLive, they have made available Rational AppScan
OnDemand, for software testing, along with Telelogic Focal
Point, an online software management tool. At IBM's size and
reach and with their treasure trove of technology anyone
interested in cloud computing will need to know what they're up
to.

High Growth Applications
At one level some might see that the difference between these
new compute and storage cloud services and the traditional
model of buying servers and putting them in the data center as
COMPUTE & STORAGE


87
only an economic choice. Of course it may be easier to purchase
and it the service could be lower cost, but to see compute and
storage only in this light would be to miss the true potential.
Let us suggest three styles of applications that can truly benefit:
very high growth applications; peaky applications and parallel
applications.

The first category of applications, very high growth
applications, and look like you won the lottery. Perhaps only one
out of a hundred or one out of a thousand applications might see
rapid adoption, but when it does the ability to rapidly scale is
critical. Consider the case of Animoto. Animoto is an online
service that allows you to create video productions simply. You
upload your photos, Powerpoint, choose a song and Animoto
creates an effects-heavy movie in minutes. When it launched on
Facebook, demand was such that it had to increase the number
of compute instances on Amazon’s EC2 from 50 to 3500 within
three days. “You could give me unlimited funding,” says Adam
Selipsky, VP at Amazon Web Services, “and I wouldn't know
how to deploy that many servers in 72 hours”
14


Peaky Applications
This class of applications is characterized by cyclic usage of
resources. Typically this might mean computation that occurs

14
Economist - October 25, 2008- A Special Report on Corporate IT
CLOUD



88
every day, but not every night; or every week or every month.
Typically in this scenario a traditional model requires that you
allocate the largest required resource, even thought it might only
be used 1/3 of the period.

An example of this type of workload is a Wall Street firm
with a nightly business process where they upload the day’s
market trading data into a storage service (in this case Amazon’s
S2), and then run proprietary risk management algorithms. The
analysis typically lasts ~10 hours during week nights and during
that period 3000 compute instances are used. That demand is an
order magnitude greater than the daytime and weekend workload
where the number of instances is closer to 300.

Parallel Applications
As part of teaching the first class on cloud computing at
Tsinghua University, Amazon donated $3000 worth of time on
the Amazon cloud. At $0.10 per hour it means you could use
one instance continuously for 3.5 years, or 100 computers for 12
days or 10,000 computers for 30 minutes. But what would you
do with 10,000 computers for 30 minutes? We’ve never been
able to ask that question before and the answers promise to be
truly exciting.

COMPUTE & STORAGE


89
There are only two areas in business computing that have
taken advantage of large numbers of independent low cost
computers. In the 80s Tandem Computers created systems that
allowed 256 computers to outperform the most powerful
mainframes of the era in a transaction processing system, which
was optimized to enable parallel execution. And beginning in
2000, Google has shown the power of using 1,000s of computers
to enable low cost, high performance search. So the challenge is
to identify (and parallelize) algorithms that are not as simple as
consumer search.

One of the chief architects of the Tandem system, Jim Gary,
in more recent years turned his attention to what he called the
“fourth paradigm” of science
15
. The first three paradigms were
experimental, theoretical and computational science. He
explained this paradigm as an evolving era in which driven by
our ability to capture a huge amount of data – whether that’s
about the human genome or the interlocking web of 300 million
people’s social graphs. Jim’s vision was that eScience could
unify theory, experiment, and simulation with a new generation
of applications specifically designed to manage, visualize and
analyze this even deeper Web of data. Based on low cost
compute and storage services Jim saw a world in which all of the

15
“The Fourth Paradigm: Data-Intensive Scientific Discovery.” Edited by Tony Hey, Stewart
Tansley and Kristin Tolle. Microsoft Research.
CLOUD



90
science literature is online, all of the science data is online, and
they interoperate with each other. Sadly in 2007 Jim disappeared
in his boat off of the coast of California. While his future
contributions will be missed – he has certainly put yet another
stake in the ground for what might be.

HP Story
HP has formed a joint cloud computing effort called the
Cloud Computing Test Bed (CCTB) that will allow researchers
from allied academic institutions and industry to test software
written for cloud infrastructure. The idea is to provide a large
cloud-computing infrastructure where people can test distributed
software at all levels of the stack, from low-level OS, storage,
and networking technologies all the way up to user-facing
applications. The CCTB will consist of six sites, each of which
will house about 1,000 to 4,000 processor cores. When those six
initial centers are ganged together into one large cloud, the
resulting machine will have a core count that should put it
somewhere in the Top 20 supercomputers list. This compute and
storage cloud uses HP hardware and software and runs Yahoo's
Apache Hadoop, an open source platform that lets users write
and run applications that process vast amounts of data.

COMPUTE & STORAGE


91
Today’s Clouds are Small
Google’s cloud contains over 500,000 servers and many see
that as large – but is it really? Just to put this into perspective
consider a few other massive computing systems. The NYSE
stock trading system processes about 10M trading transactions
per day. Visa is believed to process around 100M credit card
authorization transactions and Google about 200M searches per
day. Now let’s imagine that every mobile phone in the world
executed on transaction per day that would be over 4,000M
transactions per day.

In fact, IBM projects that by 2011, the world will be ten times
more instrumented then it was in 2006 and the number of
Internet connected devices will grow from 500M to 1 Trillion.
Indeed when you now see the iPhone and Android as merely the
“eyes and ears” of the cloud to quote Vic Gundotra of Google,
then the potential number of transactions will only grow.
Consider real time translation, turn-by-turn navigation or
determining the supply chain that brought you the lettuce in your
local market and you’ll quickly realize today’s clouds are small.

Computer Carbon Footprint
As compute & storage infrastructure grows and as our
concern for global warming increases it will turn out that for
modern technology-powered companies the carbon footprint of
CLOUD



92
the company’s computers is the new smokestack of the 21
st

Century. In this section we discuss a simple way to calculate the
Computer Carbon Footprint™ (CCF) of any company.

The CCF is computed as follows:
CCF =#Employees × SPE×TBY × PUE
avg
×CPF
avg

Where
SPE = Servers Per Employee
TBY = Technology Birth Year factor X 400 watts
PUE
avg
= Average Power Utilization Efficiency of the data centers
CPF
avg
= Average Carbon Power Footprint of the sub-region grids (lbs/watt-year)

Servers Per Employee (SPE)
The number of Servers per Employee may be a simple
measure of how much a company is powered by technology.
Interestingly, in a presentation
16
by Simon Commander of the
London Business School and Rupert Harrison of the IFS School
of Finance they measured the number servers per employee in
Brazil at 0.04 and in India at 0.02. In a book
17
published by the
World Bank in 2007 the on the economy of the country of

16
ICT adoption in developing countries: Firm-level evidence from Brazil and India; Simon
Commander (London Business School) Rupert Harrison (IFS School of Finance 1st June 2006.
17
Environmental Priorities and Poverty Reduction: A Country Environmental Analysis for
Colombia (Directions in Development) (Paperback) by Kulsum Ahmed (Author), Kulsum
Ahmed; Ernesto Sanchez-triana; Yewande Awe (Editor)
COMPUTE & STORAGE


93
Columbia the range was reported to go from 0.32 to 3.33
computers per employee.

Of course probably nothing compares to the modern large-
scale Internet companies. In 2006 the number of servers Google
deployed was estimated to be 450,000.
18
With Google
employing approximately 10,000 people in 2006 it would put
their SPE (Servers per Employee) at 45.

Year Servers Per Employee
India 2007 0.02
Brazil 2007 0.04
Columbia 2007 0.32-3.33
Google 2006 45
Figure 4.2: Servers Per Employee

Technology Birth Year (TBY)
TBY is designed to compute the approximate power
consumption of all of the servers. This is done by establishing a
base line of power consumption at 2006 at 400 watts and then if
the servers are on average older or younger, increasing or
decreasing the power consumption. We base our analysis on a
recent paper
19
authored by Jonathan G. Koomey, of Lawrence

18
Markoff, John and Hansell, Saul. Hiding in Plain Sight: Google Seeks an Expansion of Power.
The New York Times, June 14, 2006.
19
Assessing Trends in the Electrical Efficiency of Computation over Time, Submitted to IEEE
Annals of the History of Computing: August 5, 2009 Released on the web: August 17, 2009

CLOUD



94
Berkeley National Labs, Stephen Berard from Microsoft, Marla
Sanchez with Stanford University and, Henry Wong from Intel
Corporation. The paper explores the relationship between the
performance of computers and the electricity needed to deliver
that performance. As shown in the figure below, computations
per kWh grew about as fast as performance for desktop
computers starting in 1981, doubling every 1.5 years, a pace of
change in computational efficiency comparable to that from
1946 to the present.

Figure 4.3: Computation per kWH per decade

Under the above assumption that computations per server
doubling every 1.5 years then the same unit of performance
would require 4x more power 3 years earlier in 2003 and ¼ the
amount of power 3 years later, in 2009.

Technology Birth Year
COMPUTE & STORAGE


95

Figure 4.4: Technology Birth Year factor

While every server consumes a different amount of power we
will use 400 watts as the average power consumption of a server
in 2006. This will also make our estimates conservative. More
precise calculations would include the data in Figure 4.5.

Class
2000 2003 2005
Volume 183 214 218
Mid-range 423 522 638
High-end 4874 5815 12682
Figure 4.5: Weighted average power (Watts) of top 6 servers, by sales
20


Summary
Post the crash of 2001, many companies focused on reducing
the cost and as a result gave rise to the first generation of
application cloud service companies (Salesforce, Taleo,
RighNow...). While these companies have wide applicability,
the introduction of compute and storage cloud services has the

20
Koomey, J., G., 2007, Estimating Total Power Consumption by Servers in the US and the
World. February 2007. http://enterprise.amd.com/Downloads/svrpwrusecompletefinal.pdf
2003 4
2004 3
2005 2
2006 1 (Baseline)
2007 0.75
2008 0.5
2009 0.25
CLOUD



96
potential to make an even bigger impact. Again spurred by the
economic downturn of 2008 and the tightening of credit anyone
looking at deploying new hardware resources (or even upgrading
existing infrastructure) has to at least consider this new model of
computing. And just as the advent of low cost PCs spurred the
last generation of client-server applications, this new generation
of compute and storage cloud services promises again to bring a
new class of applications we’ve never seen before. Which brings
us to the subject of the next chapter – platform services for new
application development.









97

PLATFORM SERVICES for SOFTWARE
DEVELOPMENT


With the increased availability of compute and storage cloud
services just as the battle for the desktop was won by winning
the developer, so it may be in the fight to win the cloud. This
chapter is not written for the VP of R&D at a software company,
but instead is meant to give everyone else, who is not an expert a
working knowledge of the issues, tools and economics of
software development. We’ll cover the fundamental parts of the
software development lifecycle and touch on some of the new
horizontal and vertical platform cloud services which are
providing software developers with a set of tools tuned for the
CLOUD



98
world of cloud computing. We’ll see how new technologies are
bridging the worlds of developing new applications, and
managing them, which heretofore have been highly separate
disciplines.

Application Spectrum
Before we launch into discussing some of the key success
factors, it’s worthwhile noting all applications are not created
equal. While all software can be delivered as a service from the
cloud, how it is delivered will vary according to the software and
its usage.

One dimension in variation is what we’ll call the application
weight.

Figure 5.1 Application Weight

PLATFORM for DEVELOPMENT


99
By weight we mean the complexity of the business process.
Email is a very lightweight business process, meaning the way
each of us uses email is pretty much the same. This is also true
of eBay-style auctioning, Web conferencing and Internet search.
An example of a heavyweight business process might be discrete
or process manufacturing, areas that Oracle and SAP have
specialized in. The way Coca-Cola manufactures soda shares
little in common with the process Chevron uses for refining
petroleum, although both are manufacturing processes. These
processes are much more customized. No one has figured out
how to make these as customizable as changing your Yahoo
home page. As a result, it’s been the case that lightweight
business processes have been earlier to move to Model Six or
Seven.

The implications of application weight are many, first, on the
frequency of change. The lighter weight business processes
allow for a greater frequency of change. At one end of the
spectrum, manufacturing or financial applications deployed in
the traditional model have resulted in upgrades occurring once
every couple of years, often at great expense. At the other end
of the spectrum is eBay makes changes 52 times a year.

Another evaluation you’ll need to make when positioning
your offering is to map the number of users for your software
CLOUD



100
against the frequency of use. This usage spectrum is shown in
Figure 5.2.


Figure 5.2 Usage Spectrum

Corporate applications, like SAP Financial Applications,
usually have very few users, and, furthermore, those users use
the application infrequently. The closer you get to the consumer
Internet, the more dramatic the change in both the number of
users and the frequency of use and therefore on the degree of
specialization and focus for the application.

As you think through your product offering, figure out where
in these maps you’re going to fit; it will have many implications
for your product and delivery options.
PLATFORM for DEVELOPMENT


101

Multi-tenancy
One area that is hotly debated is that of a single-tenancy
versus multi-tenancy model. If you’re coming from the world of
traditional enterprise software, you only architected for single-
tenancy. Adopting Model Four or Five allows your software,
architected for traditional deployment, to be deployed as a
service in the cloud and gives the customer some, but not all, of
the flexibility they have come to expect in an on-premise
deployment. This approach does not require rewriting or
building any new software and allows you to offer the service to
the 100s or 1000s of businesses who have already bought your
software.

Of course, if you have a lightweight business process, and
you intend to write the application from scratch, you have to
consider a multi-tenancy model. In this approach, a given table
can include records from multiple tenants stored in any order; a
Tenant ID column associates every record with the appropriate
tenant. The multi-tenancy approach allows you to serve the
largest number of tenants per database server. However,
because multiple tenants share the same database tables, this
approach may incur additional development effort in the area of
security, to ensure that tenants can never access other tenants’
data.
CLOUD



102

Perhaps more important, the multi-tenancy approach forces
there to be one code line and, therefore, vastly simplifies the
development and management of the software. eBay has
maintained one code line since they were founded. Of course,
this is only possible in areas where you have lightweight
business processes, and it has exponential advantages when you
serve millions of users in a highly fluid environment.

But just when you thought it was a black-and-white decision,
let’s make it more confusing. Most of the multi-tenancy service
providers run multiple sites. Furthermore each of these
applications can be on a different revision level. This is often
done to do A-B testing of new releases, as well as a change
management technique. In the limit, if General Motors wanted
one of those sites to run only GM auctions, how is that different
than single tenancy?

Bottom line, don’t get embroiled in the religious war here—
there are numerous tradeoffs to be made, depending on whether
you’re starting with an existing application or building from
scratch; whether you have a lightweight or heavyweight business
process; and finally the margins and pricing you’ll need to be
successful in the market.
PLATFORM for DEVELOPMENT


103

Underlying this whole debate is a much more fundamental
principal that’s illustrated in Figure 5.3: dramatically reducing
the cost and improving the quality of anything starts with
standardization. If something is standardized, you can
specialize. Specialization leads to repetition and, ultimately,
computers are far better at repetition than people. If you want
the highest quality, lowest cost automobile, choose a Toyota, not
a McLaren F1. While the idea of a hand-built automobile with a
production run of fewer than 100 cars certainly has exclusivity
going for it, at $1,000,000 for the car and the need for a
McLaren mechanic, you can’t argue its low cost and high
quality. So how does this apply to software?

Figure 5.3 Standardization

If we were to take a paintbrush and walk around to every
computer that has deployed an SAP or Oracle enterprise
application, we’d end up coloring each one differently, as each
CLOUD



104
installation would be on different server and disk architectures.
Furthermore each would be deployed on a different software
stack—different versions of operating systems, databases, and
middleware. Beyond that, each deployment would have a
different change management process or disaster recovery
process implemented by different groups of people. So if you
look at Figure 5.3, every box on the right-hand side has a
different color. Over at the other end of the figure, think about
eBay, with three sites, each running the same software, hardware
and management processes. Now, all of the states in between
represent progressively greater degrees of standardization.
Every choice you make in hardware architecture, software
architecture, and management processes should be an effort to
move closer and closer to the left-hand side. With greater
degrees of standardization comes the ability to specialize, repeat,
and ultimately automate the greatest cost you have—the human
labor required to manage these systems.

Software Development Lifecycle
Few books on software engineering have been as influential
as Fred Brooks' Mythical Man Month, originally published in
June 1975. Much of the book was based on his experiences as
the development manager of IBM's OS/360, at the time, the
largest single software development project ever undertaken. In
PLATFORM for DEVELOPMENT


105
the book, he recounts that the project could not be considered a
success. He writes of flaws in the design and execution of the
system and ultimately concludes that the project was late, well
over budget, and underperformed against expectations for the
first several releases.

Brooks documented what came to be known as Brooks’ Law:
“Adding Manpower to a late software project makes it later.”
Many in the software business can testify to their personal
experiences that validate his comment. He furthermore
estimated that any software project would use a 1/3
rd
of its
scheduled time in requirements, 1/6
th
in coding, a 1/4
th
in unit
test, and a 1/4
th
in system test time before it would be ready for
release. Assuming a 24-month release cycle, you'd spend six
months in requirements, three months writing code, and the last
12 months testing; again no real surprise to the old-timers.

Brooks later published a landmark paper
21
in which he wrote,
“Not only are there no silver bullets now in view, the very nature
of software makes it unlikely that there will be any. No
inventions that will do for software productivity, reliability, and
simplicity what electronics, transistors, and large-scale
integration did for computer hardware. We cannot expect ever

21
Brooks, Fred. No Silver Bullet: Essence and Accidents of Software Engineering. IEEE
Computer, April 1987.
CLOUD



106
to see twofold gains every two years.” Throughout the past 30-
plus years, the industry has struggled to find a way to uncover
Brooks’ elusive order-of-magnitude improvement in software
productivity. Rapid prototyping, object-oriented programming,
program verification, and Software Engineering’s Capability
Maturity Model (CMM) all have been thought to be the “silver
bullet,” but, unfortunately, each attempt has come up short.

While building and deploying high quality software is
certainly no easy task, as you will see in this chapter, the
movement to delivery of software as a service is fundamentally
changing the way software is designed, developed, and tested.
Maybe this isn’t the silver bullet Dr. Brooks talks about, but you
be the judge.

Horizontal Platform Services
Once you know the kind of application you want to build it
remains to consider how you’ll build it – what your coding
framework will be. As shown in Figure 5.4 you have three large
choices: horizontal software, horizontal services, and vertical
services.
PLATFORM for DEVELOPMENT


107

Figure 5.4 Platform Cloud Services

Horizontal software refers to a framework where you may use
any of the software (open-source or traditional Model 1
software) on a compute and storage cloud service. In this model
you can choose to use the database, middleware and languages
of your choice. All of today’s application cloud service
providers, who did not have horizontal or vertical platform
services to choose from, use large amounts of horizontal
software on compute and storage services they have had to
manage.

Which brings us to horizontal services. Horizontal platform
services offer you less flexibility, but allow the cloud service
provider to manage the platform software thereby reducing cost
and complexity. Further restrictions in flexibility (e.g., definition
of a particular database schema) result in more vertical platform
CLOUD



108
services in which you’ll further trade off flexibility, for greater
productivity and lower cost assuming you’re application maps
well to these vertical platform services.

In order to understand the advantages of these platform
services it’s useful to further understand the traditional
application environment. Figure 5.5 shows the traditional silos
between the group that writes the original application, the group
which has had to port and test it in multiple operating system
and database environments and finally the operations
management group which has to manage the application after
it’s delivered.


Figure 5.5 Traditional Application Development & Deployment

What this new generation of platform services is attempting
to do is to merge these worlds together since now the builder of
PLATFORM for DEVELOPMENT


109
the software is no different than the people who have to operate
the software. As an example these platform services provide:

• Maintain a current level of the underlying software
• Automatically replace a failed server
• One-button click to replicate production application
for testing and staging
• Automated change management – from code check in
to production
• Add capacity when load thresholds trigger.

Engine Yard is a good example of providing horizontal
platform services to the Ruby-on-Rails community. Google’s
App Engine, while it has a number of programming model
restrictions, does give you access to the highly scalability of
Google’s compute and storage services. More importantly,
versus a traditional LAMP stack deployment frees you from the
operations management hassles like disk space management,
security patch management and backup and recovery. Finally,
Microsoft, who ruled the client-server programming model, has
introduced Azure to provide an easy on ramp for the millions of
Microsoft programmers around the world to build next
generation application cloud services using many of their
familiar tools.

CLOUD



110
Microsoft Azure Story
In a widely publicized email to top executives, dated Oct. 30,
2005, Bill Gates, founder and then Chairman of Microsoft urged
company leaders to “act quickly and decisively” to move further
into the field of offering software as services, in order to best
formidable competitors. “This coming 'services wave' will be
very disruptive,” Gates wrote. “We have competitors who will
seize on these approaches and challenge us—still, the
opportunity to lead is very clear.” Gates compares the push
toward such services, which range from online business software
offerings to free Web-based email, to the changes he saw nearly
a decade ago. Then he wrote a now-famous memo, The Internet
Tidal Wave, which prompted a massive shift at Microsoft toward
Internet-based technology. “The next sea change is upon us,”
Gates wrote. While Gates might have left day-to-day
operational responsibilities, there are key people to carry on his
vision. First and foremost is Chief Software Architect, Ray
Ozzie. In a recent interview Ozzie said, “We’re going to look
back at this era and wonder how we did without this other kind
of computer in the cloud.”

Windows Azure seeks to provide the Microsoft Visual Studio
programmer with compute and storage services similar to those
provided by Amazon as well as SQL and .Net services with
PLATFORM for DEVELOPMENT


111
integrated operations management. As an example the SQL
Azure service provides availability monitoring on five minute
intervals as well as updating lower level system software as
required.

Google AppEngine Story
In mid-2008 Google introduced App Engine. While with
Google Apps they provide office applications in the cloud, App
Engine is their first attempt to provide a platform for others to
build cloud-based applications. Speaking at App Engine
Campfire One in April 2008, Kevin Gibbs, App Engine Tech
Lead said, “App Engine is different than a lot of other things out
there. App Engine is not a grid computing solution. We don't
run arbitrary compute jobs. We also don't give you a raw virtual
machine. Instead, we provide a way for you to package up your
code, specify how you want it to run in response to requests, and
then we run and serve it for you. You don't reserve resources, or
machines, or RAM or a number of CPUs, or anything like that.
It's a fluid system that runs your code in response to load and
demand.”

Ultimately, App Engine’s goal is to provide a simpler
alternative to the traditional LAMP stack. By providing a more
vertical (more constrained) offering the tradeoff will hopefully
CLOUD



112
be greater productivity at the expense of flexibility. If you want
flexibility you can always go buy your own computers.

App Engine starts by writing server side code in Python. You
can make use of numerous Python modules and libraries using
any Web framework. The database is not SQL instead it is
based on Bigtable, the system used for Google.com. Bigtable is
a fault-tolerant, distributed system that can span tens of
thousands of hard disks on thousands of machines, making all of
them appear to be a single storage table. It is designed to move
data around and restructure the system automatically to account
for hotspots and increased storage. Unlike SQL, Bigtable is
schema-less, meaning it can support arbitrary new properties or
columns, which can be created as you code. One feature you
won’t find is joins, since they are typically a source of
performance problems in a distributed system.

As we know, not only does there need to be a programming
environment, but there also needs to be an operations
environment. Toward that end an administration console is
provided to give you access to the application. While there are
some early applications being demonstrated, it is still early days.

PLATFORM for DEVELOPMENT


113
Vertical Platform Services
Further down the spectrum of decreasing programming model
flexibility for increased productivity and decreasing cost to
manage are the vertical platform services. Vertical platform
services assume you are building a specialized usage of the
wider platform provided by the vertical platform cloud service.
This typically means you’re leveraging at least the data model
and hopefully some of the higher level abstractions provided by
that data model.

Good examples of vertical services include Netsuite’s NS-
BOS, Salesforce.com’s force.com and Facebook’s platform.
Netsuite seeks to allow application builders who want to
specialize their ERP oriented platform for more specific markets.
Salesforce with force.com seeks to provide not only the
technology, but also a marketplace (AppExchange) to encourage
new CRM applications. Finally with so many users of Facebook,
by opening up the platform (most notably Facebook Connect)
can leverage a wider development community to extend their
services far faster than they can with their captive R&D group.

Salesforce.com Story
In traditional business software, Salesforce.com has led the
way in turning their platform into a service others can use.
AppExchange was launched in late 2005, ironically almost at the
CLOUD



114
same time that Oracle acquired Siebel. At the time, the objective
was to enable customers and partners to customize the
Salesforce.com application and extend it. There was no firm
plan to extend the core application platform beyond its forms-
driven database model. Phil Wainewright, who blogs about
SaaS for ZD Net, writes that two events changed
Salesforce.com’s approach. The first was an attempt to use the
platform to write a quote application, which required
transactional integrity. The other key event was that developers
began to build applications that were not the conventional forms-
based apps. The result, according to Phil, was the creation of the
Apex programming language ultimately leading to a complete
platform for building application cloud services. Apex provides
Force.com platform with a procedural language so that users can
go beyond declarative programming to write code where needed.

You can use this language to create triggers, which fire when
any user performs certain data operations. Apex can also create
classes, which can be called from other classes and triggers,
enabling productive code reuse and standardization. A year after
the introduction of Apex, Force.com platform added
Visualforce. Visualforce is a framework to build custom user
interfaces. The Visualforce framework includes a tag-based
markup language, similar to HTML. Underlying the Force.com
PLATFORM for DEVELOPMENT


115
platform is a multi-tenant database. The Force.com database
differs from relational databases in the way record relationships
are implemented. Instead of developers’ having to deal with
keys, foreign keys, primary keys and defining relationships in
terms of these keys, the database deals in terms of relationship
fields. A relationship field stores the ID of the parent record in a
relationship, as well as optionally providing user interface
representations in both the parent and child records. The
Force.com platform supports two query languages:

• Salesforce Object Search Language (SOSL) is a language
for searching all across all persisted objects.
• Salesforce Object Query Language (SOQL) is a query-
only language. While similar to SQL in some ways, it's
an object query language that uses relationships, not
joins, for a more intuitive navigation of data.

Of course, the real test is the types of applications that can be
built and delivered using the salesforce.com cloud. Two
interesting case studies are Riskonnect and Stakeware. Bob
Morrell founded Riskonnect in July 2007. He’s had a long
history in risk management software beginning with Risk
Laboratory, which he started in 1994.

Riskonnect’s software, launched in 2008, helps people
visualize and understand risk at very high levels. “This is
CLOUD



116
essential because too many consultants only use complicated
calculations. But many people don’t need that,” Morrell said.
Riskonnect allows users to drag and drop information to
compare risks visually. An example would be if one factory
goes down, how would other factories and the supply chain
respond.

Natan Zaidenweber founded StakeWare, another force.com
application, in 2004. Zaidenweber has over 15 years of
experience in sustainability as a researcher, lecturer, and
manager working in projects for The World Bank, The Mexican
Government, and Shell Oil. StakeWare enables corporations to
gather, organize, and analyze stakeholder data, identify key
issues and related indicators that are relevant and material to
include in the standard G3 sustainability report. The software
also includes the ability to monitor and analyze emissions to
help provide appropriate trade-offs when lowering the carbon
footprint as well as the management of sustainability policies
across supply-chain.

Each of these examples is just further proof of what Marc
Benioff said: "Software developers from a developing country
can build just as great an application on our platform as
somebody who lives in Palo Alto.”
PLATFORM for DEVELOPMENT


117

Facebook Story
In February 2009, a little more than a month after announcing
it had 150 million active Facebookers, Dave Morin, who runs
the application platform team, twittered the number was now
175 million. That kind of growth and scale can in part be
attributed to Facebook opening up its platform and providing a
full suite of APIs—including a network protocol, a database
query language, and a text markup language—that allow
applications to integrate tightly with the Facebook user
experience. Their most recent effort has been to bring the power
of Facebook to other applications by providing the ability to:

• Connect any existing Facebook account and information
with any application.
• Connect and find friends who also use the application.
• Share information and actions on the application with
their Facebook friends.

This is a particularly valuable service since in the cloud many
people will not represent, nor are incented to represent
themselves, with their true identity. How many times have you
signed up for a site or a newsletter with fake identity? On the
other hand it’s difficult to have a fake Facebook identity, after
all unless you have no friends, it’ll be difficult to pretend you’re
Angelina or Brad. Of course not only is basic profile
CLOUD



118
information available but also the profile picture, friends, photos,
events, groups, and more.

With access to the social graph, an application can
dynamically show which Facebook friends already have
accounts on the application. As an individual moves around the
Web, their privacy settings will follow them, ensuring that users'
information and privacy rules are always up to date. For
example, if a person changes their profile picture, or removes a
friend connection, this is automatically updated in the
application. In addition, the individual can control who can see
what pieces of their information—the same rules that they set on
Facebook can be applied through your application. If you’re
interested in seeing early examples of applications using the
Facebook platform take a look at TechCrunch, Citysearch,
VLane, and Howcast.

Interestingly Facebook has published a set of principles for
their applications. Some of these are clearly unique to their
applications—others are just good ideas.

Applications should be meaningful.
• Social – Helps people interact and communicate more
effectively by using information from the social graph
PLATFORM for DEVELOPMENT


119
• Useful – Delivers value to people by addressing real
world needs, from entertainment to practical tasks
• Expressive – Enables people to share more about who
they are and about the world around them.
• Engaging – Provides a deep experience that people want
to come back to regularly

Applications should be trustworthy.
• Secure – Protects user data and honors privacy choices
for everyone across the social graph
• Respectful – Values individual attention and honors their
intentions in communications and actions
• Transparent – Explains how features will work and how
they won't work, especially in triggering person-to-
person communications

Applications should be well designed.
• Clean – Designed to be intuitive, easy to use and free of
mistakes
• Fast – Achieves low latency while scaled to handle
individual demand
• Robust – Maintains reliable uptime and minimizes error
rates

With such a large population of people on Facebook, a
successful application on Facebook can grow to millions of
people in a short time. A case in point is the story of iLike.
When iLike launched 10,000 people joined in the first 12 hours.
CLOUD



120
By the end of the first day they had 50,000 people. In less than a
year, iLike announced they had passed three million and were
growing at a rate of 300,000 per day. Now if we all were so
lucky.

NetSuite Story
In 2005, NetSuite opened its architecture to partners and
outside developers to further its reach through its NetSuite
Business Operating System (NSBOS), a public cloud
development environment that includes a set of Application
Program Interfaces (APIs), a mechanism to distribute
applications, and the NetSuite application itself. To use Zach
Nelson’s words: “By using the cloud, we can now economically
reach the Fortune 5,000,000".

As a more vertically focused cloud service, NSBOS targets
building out the core NetSuite offering with additional
specialized functionality beyond the services delivered by the
core application. The resulting specialized application and data
reside inside the NetSuite platform and share database and server
resources within the core solution. This tight integration enables
the platform and applications to remain in sync as the underlying
NetSuite application is upgraded.

PLATFORM for DEVELOPMENT


121
The NSBOS APIs, called SuiteFlex, give application
developers data storage flexibility. Developers can create
custom forms to give users access to specific fields, and create
completely new custom record types for objects unique to a
specialized application. SuiteScript, the JavaScript API, allows
developers to add workflow to records and automate business
processes for target industries with full transactional integrity.
Integration with external systems is supported through
SuiteTalk. SuiteTalk uses industry standard SOAP and XML
messaging to communicate with external applications. The
platform also allows developers to build Web services
integrations using .Net, Java or PHP. Finally, the resulting
application can be packaged using SuiteBundler, which enables
members of the SuiteCloud Developer Network to distribute
their applications on SuiteApp.com.

As a result of this strategy NetSuite has focused on
specialized developers. “Rather than cast a wide net, we target
partners with the industry experience to create compelling
applications,” says Guido Haarmans, NetSuite VP of Developer
Programs. A couple of examples of these specialized
applications are those provided by Roostock Software and SPS
Commerce. Pat Garrehy, a veteran of the manufacturing and
Material Requirements Planning (MRP) software, founded
Rootstock. His previous company developed conventional
CLOUD



122
Model One ERP/MRP software that was used by companies like
Lockheed and Solectron. In 2008, Garrehy got the itch to return
to the MRP business. He recognized the rise of cloud computing
and rather than replicate fundamental ERP functionality he
chose to build on NetSuite. With NetSuite providing the
underlying cloud services, they built a multi-plant MRP
application, in just nine months.

SPS Commerce, a leader in electronic data interchange (EDI)
supply chain developed a specialized application that allows
smaller suppliers to tap into their network of over 1,300 retailers,
thereby allowing a small supplier to quickly and inexpensively
meet the complex EDI requirements. Circle of Friends and Mad
Croc, both small CPG manufacturers, benefited. Neither could
afford traditional ERP and EDI applications and the IT resources
necessary to keep them running. But without an EDI solution
they were limited in which retailers they could work with. By
adopting the vertical application cloud service they could
quickly and inexpensively manage their entire trading
relationships, from inventory to shipping and invoicing.

Testing
As Fred Brooks noted, most of software development time
and cost involves testing. So how does moving to an on-demand
PLATFORM for DEVELOPMENT


123
model change anything as regards testing and QA? First of all,
moving to the model reduces the number of variations and
creates greater standardization. Whether that’s a standardized
delivery model in Model Four or Five, or using one line of code
when you move to Models Six or Seven, reducing the degree of
freedom reduces the number of overall tests that have to be
developed and executed. Now, instead of trying to create and
manage tests for this application release (and the three previous
releases) and insure that they work on Oracle, IBM and MySQL
databases on HP, IBM or Sun hardware, the QA team can limit
their focus to one line of code (hopefully), one database, and one
hardware platform.

Furthermore, the model lets you begin to leverage tests of
specific customer configurations and execute those tests very
early in the development cycle. So, rather than wonder whether
an upgrade will work, you’ll know, since the application will
have already been tested in the development labs. Collapsing
the supply chain between the customer and the producers of the
software has even further economic benefits. Rather than
employ an army of test developers in your engineering
organization, you can use the actual customer systems to
develop real-world tests. The tests are more representative and
come at a much lower cost.

CLOUD



124
But there are also major new responsibilities and issues. For
instance, if you introduce a bug into production, 100 percent of
your customers experience that bug immediately. Think about
that. Concur co-founder Michael Hilton said, “For Concur, that
means over six thousand companies and over four million end
users are affected by any defect.” It has fundamental impacts on
how you think about quality assurance. But there’s a flip side as
well—you can fix the issue and 100 percent of your customers
have that issue resolved—no more waiting for the slow trickle of
customers to download and install the patch.”

Of course, just as we will see in the next chapter on
operations, the testing/QA function will need to re-engineer for
the new reality of a release a week or a release a month, not a
release every 24 months. Again, as Hilton points out, “This
fundamentally changes how you develop. You end up releasing
things in smaller chunks, so users can easily digest them. And
you end up releasing more often. At Concur, we do meaningful
functional updates to our service every single month. That
means that, every month, our users are getting new innovations
and benefits without having to do anything. It forces you to care
a lot more about the user experience, ease of use and change
management. Lots of things end up being optional, meaning
PLATFORM for DEVELOPMENT


125
that, if the feature is big enough, customers don’t have to deal
with it right away if they don’t want to.”

Customization
One of the major objections to the SaaS/application cloud
service model has been the belief that there is no ability to
customize. “You can have any color you want as long as it’s
black,” was the famous Henry Ford quote often used by people
describing early offerings.

While this subject is almost nonexistent from a consumer
perspective, any conversation with corporate IT will inevitably
lead to the discussion of how to customize the application. As
Oracle On Demand was blessed with a natural channel to
medium and large-scale businesses, it remained challenged on
how to address this question. Some engineers would try to be
PC (politically correct) and say, “We don’t allow for
customization, we allow for configuration.” Or others would
say, “We don’t support extensions, we support configurations.”
The conversation was confusing. Marian Szefler, today Senior
VP of R&D at Risk Management Solutions, architected an
innovative approach while at Oracle that changed the argument
from techno-babble to a business decision.

CLOUD



126
In 2000, Marian defined the CEMLI framework. CEMLI
stood for Configuration, Extension, Modification, Localization
and Integration, and described 17 classes of reason why
someone other than Oracle R&D might need to add software to
an Oracle application.


Figure 5.6 CEMLI Framework

The framework shown in Figure 5.6 lists the 17 classes and
is, of course, unique to Oracle applications. You should take
away a few ideas from this. First, by creating these classes, for
the first time, one could have a precise conversation. “CEMLIs”
could be counted, grouped, and discussed by both the customer
PLATFORM for DEVELOPMENT


127
and Oracle—the framework eliminated the politically correct
dialogue that preceded it. Second, not all CEMLIs were created
equal. Adding a new report was far less invasive than adding a
trigger, for example. The classes were ranked by how “toxic”
they were, meaning the likelihood that a mistake in this class
would cause data corruption or a down system. In fact, as you
got to higher degrees of “toxicity”, you were coming closer and
closer to making fundamental modifications to the software,
something you want to avoid at all costs. Finally, the CEMLI
framework created an environment where the degree of
customization became a business decision, not a technical or
religious one. By isolating CEMLIs and separately pricing the
management of these, the customer could choose how much they
wanted to spend, for what degree of uniqueness. There were
cases at Oracle where the cost to manage the CEMLIs was larger
than managing the entire Oracle application.

Fred Magner, former CIO of Unocal, used the cost to manage
CEMLIs to move the number of CEMLIs down from 1500 to
300 over the period of 18 months. The model created a
framework for the business to justify why the particular CEMLI
was worth the extra expense. So now the CIO could ask the
business, “Do you want to spend X for the uniqueness, or can we
use the standard product and save money?” This is no different
than in the real world, where standard products, like Toyota
CLOUD



128
Corollas, are far lower cost (and higher reliability) than custom
crafted cars like a McLaren F1. The cost for servicing the “one”
is many more times than the cost of servicing the “many.”

Anyone in enterprise software has faced these challenges.
Taleo has addressed the customization challenge by creating a
single line of code for all customers while baking in the ability
for customers to do extreme configurations. This approach
provides customers the ability to avoid the customization version
lock trap while still enabling customers to meet their unique
business requirements. Taleo CEO Michael Gregoire said, “This
approach did result in us having to pass on some very large
opportunities in our formative years when prospects insisted on a
customized approach. However, years later, our approach has
been validated, when some of these very same prospects have
returned to look at Taleo to replace those same customized
solutions purchased earlier from other vendors.”

Concur’s Michael Hilton sees the investment in architecting a
solution as one reason traditional R&D and consulting are now
connected. Concur created a model that allows them to easily
move customizations into core code. He said, “It’s just part of
our development model. When you think about traditional
enterprise software companies, a lot of times the teams that write
PLATFORM for DEVELOPMENT


129
customizations (typically consultants in the field) are completely
separate from the core R&D teams. In our model, that doesn’t
work. Both teams need to be working in the same organization,
utilizing the same tools and processes, so that everything scales,
migrates, and works together seamlessly. At Concur, we built
out a fairly complex architecture to deal with customizations. It
took us a couple of years to really get it right, but the benefits are
extraordinary. Our large clients can have their unique needs
meet, while still operating in our multi-tenant environment.”

Traditional vs. Cloud Development
But, let’s get back to R&D. Perhaps no one is better qualified
to talk about the differences between the traditional development
and development in Model Six than Evan Goldberg. He speaks
from an excellent vantage point, having spent eight years with
Oracle, followed by ten years with NetSuite, as Founder,
Chairman, and CTO.

In a lecture at Stanford University, Goldberg pointed out
there are numerous product design benefits to Model Six. First,
it is much easier to talk to the customers. As you’re delivering
the service, there is no “middleman” between the producer of the
software and the consumer. What’s more, since the software can
be instrumented, you can tell what features are being used, what
features aren’t. Amazon.com takes special advantage of
CLOUD



130
designing an operational environment that allows them to
experiment with new features and, at the same time, compare
with the old system. They can use data to determine what
features to keep and what to retire. And, because there can be
universal adoption of innovations, as a designer, you can build
on top of previous innovations rather than worry about which
customer has which features.

The immediate feedback loop afforded by the model allows
the software engineers to, as Goldberg put it, “Kill Bugs Dead!”
There is instant gratification to solving complex problems,
versus the traditional model, where it may be months before a
particular patch can make it into a production system. With an
ability to focus on single version, software engineers can put all
of their energies into that release, versus working on back
porting to multiple releases on multiple platforms. Finally,
given that installation is occurring on the companies’ computers,
the engineering staff doesn’t need to spend a lot of time on
simplifying installation, something, Goldberg said, “they were
never good at.”

The last group Goldberg addressed in his Stanford lecture
was the QA organization. As we’ve already pointed out,
delivering software as a cloud service allows for testing through
PLATFORM for DEVELOPMENT


131
actual customer usage. The ability to monitor real-time
performance and develop and inventory tests based on actual
usage are significant advantages over the traditional model. The
challenge that QA shares with the other disciplines is that each
release of software is happening much more quickly. In the case
of NetSuite, the average has been over three releases per year.
With such a rapid rate of change, we can’t depend solely on
human power so all aspects of engineering are going to have to
automate.

Goldberg ended his comments by putting up a slide that said:
“Operations: Welcome to the New World.” He noted that, for a
VP of R&D in a traditional software company, worrying about
the operations of your software was a distant issue. There was
always the support organization and the customers own
operational people that stood between you and any issue. But in
any of Models Four through Seven, that idea is a distant
memory.

Summary
We’ve attempted to highlight some of the more important
differences between traditional software development and the
new world of developing software for cloud deployment.
Whether it’s database architecture, how you architect for
CLOUD



132
customization, or how you’re going to have to re-think testing
and software release, it’s a new world.

In our next chapter, we’ll spend time on the key success
factors in the management and operations of software. As Evan
says, “As the VP of R&D, you’re going to need to understand
how to manage the security, performance, problems, change and,
perhaps most importantly, the availability of your software.”









133

PLATFORM SERVICES for
OPERATIONS


Repeat after me, the key to excellence in anything is
specialization and repetition. This is true for all human
endeavors. Swimmers swim thousands of laps to perfect a
stroke, a student recites multiplication tables to learn to multiply,
and a great surgeon performs hundreds of surgeries to master her
craft. This is no different in the world of managing software
applications. Today, an application cloud service provider
manages hundreds of systems and therefore has to repeat
numerous processes multiple times. Don Haig, vice president
for Oracle, cites an example of where this repetition ultimately
CLOUD



134
helped his customers, the users of Oracle applications. “Take
software updates, an event that most organizations have to
manage with some frequency and yet is an annoying distraction
from their core business. CIOs really don't like to do updates,
but understand they need to stay current on the latest release of a
vendor's software. When a customer is managing their own
systems, they will do this upgrade once and only once. In
Oracle On Demand, upgrades have been repeated hundreds of
times per year.”

Upon hearing this example, Navy Captain Brian Kelly, CIO,
TRICARE Military Health Services, who was trained as a
neurosurgeon, said, “You don't need to explain this further. We
have an expression in my business—‘You don't want to be the
first guy to get your cabbage worked on’.” Yes, but in most IT
departments today, complex surgeries are done for the first time
by people who read the book the night before.

Consider the following research in the medical profession in
light of the complexities of managing IT infrastructure. A
simple question posed to surgeons, the research shows, can often
separate the talented from the average: How often do you do
this? Research has established that busy hospitals generally
deliver better care. But recent work adds more detail, citing the
PLATFORM for OPERATIONS


135
specific risks patients face in the hands of low-volume surgeons,
who don't frequently and regularly perform certain procedures.
“It is now a very legitimate question to ask a surgeon: How
much experience with a procedure do you have? What are your
complication rates?” Tufts Medical Center's surgeon-in-chief,
Dr. William Mackey said. ”It’s something every surgeon should
keep track of and show patients if requested.” Do we ask the
same of our operations staff? How many times have the
operations staff upgraded from Release 1 to Release 2?

Some surgeons are convinced that measuring patient volume
alone is too crude a measure for grading surgeons. But
increasingly, the profession sees constant repetition as key to
mastering complex surgeries. “The more often I do a procedure,
the more I fall into a standard routine,” said Mackey, a vascular
surgeon who performs frequent carotid endarterectomies, a
delicate procedure to clear a key artery of blockages. “Then, if I
encounter something out of the ordinary, I'm quicker to
recognize it and correct it.” Many studies of recovery from
computer outages show the outage is often extended because the
operator has seen something out of the ordinary. Since it may be
the first time they've seen the issue, they don't know how to
react.

CLOUD



136
So what’s a high degree of repetition? Dr. Mark S. Litwin, a
University of California at Los Angeles urology professor,
authored a study where high volume meant 40 or more
procedures annually. Consider a corporation's system, network,
and database and application operations teams. How many
times a week have they performed a particular procedure? Is the
procedure even written down? Can it be repeated?

The good news is that if a process can be repeated 100s of
times it can be automated. And computers are far better and
lower cost than people. If we know the performance
management process, or the upgrade process, and we’ve
repeated it a hundred times, then why not program computers to
automate the process?

To better understand what it means to automate key processes
consider a simpler problem. If you went bowling ten years ago,
rest assured, the first question, after “Where do I get a beer?”
was likely, “Who’s going to keep score? Who knows how to
score?” You can be certain the same conversation was occurring
in all ten bowling lanes, each with a different answer to the
question. So, while all ten lanes were bowling, did a score of
200 on Lane 1 compare to a 180 on Lane 2? It was anyone’s
guess, since scoring was left to whoever said they remembered
PLATFORM for OPERATIONS


137
how. Ten years later, the story is quite different. Now, if you go
into a high-tech bowling alley, or play Wii Bowling, you enter
your name, and the computers score and tell you whether you
get to bowl again. You can be guaranteed that a score of 200 in
Lane 2 and a score of 180 in Lane 3 are scored identically. It
might not improve your bowling, but you can make sure that the
scoring process is being implemented consistently. Consistency
and standardization are being enforced through automation.

In the remainder of this chapter, we’ll go through some key
processes in security, availability, performance, and problem and
change management the VP of Operations, or VP of R&D will
need to focus on.

Change Management
One of the advantages of Models Four through Seven is the
increased simplicity of moving customers forward to the latest
release. From an R&D point of view, this is good, since your
developers can focus on one code thread versus being forced to
pay attention to multiple threads and platforms. And from the
customer’s point of view, they get to receive the benefit of new
R&D, versus putting it on the shelf and avoiding the upgrade at
all cost.

CLOUD



138
We should also note that upgrades or changes to enterprise
business software have always been an expensive and difficult
proposition. Two Northern Illinois professors documented
22
the
cost and challenges of upgrading Enterprise Resource Planning
(ERP) applications. Factors that range from weak project
leadership to minimal employee product training have resulted
in many ERP projects being delivered late and over budget, with
costs that were on average 25 percent over their original
budgeted amount
23
. The cost of a typical ERP implementation
in a Fortune 500 company was estimated as between $40M and
$240M, according to AMR Research.

Change management for applications, platform and system
software sits at the nexus of the cost and reliability of delivering
applications as a cloud service – there may be no more important
process. Consider these four fundamental principles when
designing your process:

1. Limit the locus of change
2. Standardize the frequency of change
3. Architect the software for change (roll-in & roll-back)
4. Instrument the environment.

22
ERP II: best practices for successfully implementing an ERP upgrade. Communications of the
ACM, 2006, Volume 49, Number 3, pages 105-109.
23
Peterson, W.J., Gelman, L., and Cooke, D.P. ERP Trends. The Conference Board Report,
2001.
PLATFORM for OPERATIONS


139

And for the key process From-Software–Change-Initiated-
Until-Completed measure the duration in minutes. As you are
able to improve the predictability and improve the speed you’ll
not only reduce cost, but also increase the reliability.

eBay Story
If information moves along an information superhighway,
then Lynn Reedy is a master traffic engineer. As the Senior VP
of Product, Development and Architecture at eBay, Reedy led
the group that was responsible for all of the initiatives that
pertained to the company’s work in software development,
product specification, user interface design, usability testing,
architecture and release management.

Reedy has some good counsel for anyone building software
that is delivered on demand. First, she says to insure a high
availability application; don’t engineer for one big box, instead
componentize the application so you can deploy on lots of little
boxes. Furthermore, make sure you’re not dependent on one
large database but find a way to segment and componentize.

eBay architected a system that was based on the idea that 100
percent of the code base was sent to all of the pools of resources.
A pool is a set of computers dedicated to a particular function.
CLOUD



140
For example, there is a “sell your item” pool and a “My eBay”
pool. The pooling architecture allowed change to be made
gradually. Even within a pool, a new feature would be added to
one box, then 25 percent of the pool and finally 100 percent of
the pool.

eBay architected the system for change. And remember,
when you make a change, you need to do this while maintaining
availability of the service. As a result, eBay architected the
software so that code could not only be added, it could also be
rolled back. Software needed to be both forward and backward
compatible. As a result, all database changes resulted in fields
being added. eBay’s change management process, referred to as
the eBay train. With the eBay Train approach, when software
development was done for a new feature, the feature was
released to QA. QA tested the feature for two weeks. Once it
was scheduled for a train (along with other features), the feature
was subject to a battery of over 100,000 regression tests.

For large projects, the new threads were forced to merge and,
thereby, pick up the changes every two or three weeks, with new
software headed out on the next train. This put the impetus on
the new development to resolve problems or conflicts with
features coming from other organizations.
PLATFORM for OPERATIONS


141

eBay knew that, just depending on the quality of people and
process for high productivity and reliability would not be
enough. As a result, the company invested in numerous pieces
of infrastructure software. At the top of the list was the central
application logging (CAL) facility. CAL is both software and an
engineering practice to fully instrument the application. Any
exception within the system would show up as an error on the
CAL graphical user interface (GUI). These components would
be seen as a “red pool.” Exceptions could be in the code,
network, or data layer. Before CAL, eBay personnel took weeks
to understand why a component “threw an exception.” With
CAL, resolution took minutes.

eBay’s second major investment was in automated rollout and
rollback. With over 200 features being worked on at any one
time, and with an average of 15 features per train, eBay’s usage
of automated rollout and rollback were critical. Running a
system that requires 7x24x365 availability, even in the event of
earthquakes, tornadoes, or hurricanes is a day-one requirement.
As a result eBay maintains at least three data centers in the
United States. Some people have deployed fail-over sites that
run “dark” and wonder, in the event of a failure, if the systems
and processes could handle the failover. Instead, eBay runs
CLOUD



142
every site hot and limits each site to no more than two-thirds of
the overall traffic.

Reedy offers three things that every VP of Operations and
R&D should focus on: right culture, right processes, and right
metrics. Right culture means a culture of people who find a way
to “take the hill,” who think of ways to succeed, not fail. Right
process means having a daily, repeatable process that everyone
can do over and over again—what works for surgeons should
work for IT. And finally, right metrics. Metrics tell you where
you are and where you want to be. Making this as obvious to the
operations team as earnings per share was to the business unit
led to a standard of excellence—few can argue with Reedy’s
success.

Availability Management
Most people consider the most reliable, available system in
the world to be the U.S. telephone system. We often refer to
“Dial-tone” reliability to describe the highest level of
availability. On April 6, 1992, the FCC required telephone
service providers to begin reporting outages of 30 minutes or
more, which potentially affected 50,000 customers. Recently the
standard has been raised to both report on outages affecting
30,000 customers as well as to report on any outages affecting
PLATFORM for OPERATIONS


143
major airports, 911, nuclear power plants, major military
installations and key government facilities, regardless of the
number of customers/lines affected.

The results may surprise you. For the past 10 years the data
has been collected. These are the summary statistics.

Figure 6.1 Outage Profile

CLOUD



144
These outages were caused by everything from Heisenbugs to
human error to heavy rain. In Jim Gray’s landmark paper
24
he
analyzed a set of Tandem NonStop systems and determined that
the reasons why systems stopped fell into four large categories,
which operator/human error being the majority. In a more recent
study
25
of more than 500 component failures and dozens of user-
visible failures in three large-scale Internet services, it was
observed that (1) operator error was the leading cause of failure
in two of the three services studied, (2) operator error was the
largest contributor to time to repair in two of the three services,
and (3) configuration errors were the largest category of operator
errors. Operators may face such difficulties because computer
designers and programmers have frequently sacrificed ease of
use in the quest for better performance. Database software, for
example, can require a full-time staff of trained administrators to
manage it. Ironically, because hardware and software have
grown cheaper over time, operator salaries are now often the
biggest expense in running complex Internet sites.

The experience in the field leads to forming a handful of
fundamental principles in availability management:

24
Why do computers stop, and what can be done about it? Tandem Technical Report TR-85.7
www.hpl.hp.com/techreports/tandem/TR-85.7.pdf
25
Why do Internet services fail, and what can be done about it? David Oppenheimer, Archana
Ganapathi, and David A. Patterson; University of California at Berkeley, EECS Computer
Science Division, Berkeley, CA, http://roc.cs.berkeley.edu/papers/usits03.pdf
PLATFORM for OPERATIONS


145

1. Develop a high speed, predictable recovery process:
problems are going to happen, so design systems that
recover quickly.
2. Instrument to pinpoint the sources of faults
3. Build systems that support an “undo” function, so
operators can correct their mistakes.
4. Standardize the key management processes and take
operations people out of the loop

In the end the key availability management process is From-
The-Time-The-Failure-Is-Detected-Until-The-Time-The
Application-Is-Live is perhaps the most important process.
Furthermore as the recovery time is made predictable and
reduced not only will you have happier customers, but the cost
of providing high reliability will go down.

Performance Management
Your customers expect you to provide them with application
access at an acceptable level of performance. Performance and
availability management are related. If you expect that result in
2 seconds and it takes 200 seconds does it really matter if the
application is slow or down. Performance management starts
with the difficult task of workload characterization. Every
application has some key queries, transaction flows, searches,
CLOUD



146
etc. which characterize that application. Once identified these
workload characteristics can be monitored. Perhaps the most
famous one you see every day is in the upper right hand corner
of every Google Query.

Now detecting a slow application is an exercise in profiling
the performance of the application and then seeing when the
application is performing outside of those norms. A good
example of a company totally dedicated to this is Keynote that
allows you to monitor the performance of your web application
from points around the globe. Of course once there is evidence
of a slow application then we’re back to using the change
management processes we discussed earlier in the chapter.

Security Management
Before enterprises can reap the benefits of on-demand
software, providers will have to convince IT managers and CIOs
that the services they offer are reliable and, perhaps more
important, secure. For many, the push to put information in the
cloud raises the specter of information breaches. While a
discussion about securing online applications could take the
remainder of the book, we’re going to highlight a couple of
PLATFORM for OPERATIONS


147
specific areas you’ll need to think about to insure the
applications are secure.

• Authentication. Whether thru simple password
management, all the way up to two-factor authentication
with retinal eye scan all security mechanisms begin with
knowing who you are – authentically.

• Authorization. Authorization mechanisms ensure
authentication individuals have access to authorized data
per a security policy. The simpler the definition of the
security policy and the clear evidence that it is
implemented insure the highest degree of trust.

• Audit. At the core of all security management is auditing.
Audit considerations include the process to access audit
logs, retention durations and evidence of tamper
proofing.

• Encryption. When physical data protection offered by the
co-location facility is not possible (e.g., on a tape,
removable disk, network) encryption becomes the service
of choice. Encyrption also solves problems of protection
when data leaves one application domain and enters
another (e.g., PIN encryption).
CLOUD



148

In the security management area there are numerous
standards. One of the most recognized commercial standards is
the Statement on Auditing Standards No. 70 (SAS 70). A SAS
70 audit is performed by an independent auditor and results in a
SAS 70 report. One should be mindful that a SAS 70 report only
documents the internal control practices of an organization,
without offering any judgment as to whether they are
satisfactory. Due diligence therefore requires that you not only
request an SAS 70 report from a prospective provider, but that
you examine it thoroughly to determine whether the provider is
able to comply with your own internal standards for privacy,
data security, and so on.

Care Rehab Story
“For Care Rehab, a medical device manufacturer, security
was an important consideration when the company was
evaluating Salesforce.com,” said Ed Barrett, vice president at the
200-person company. The company, which makes traction and
electrotherapy devices used by physical therapy clinics, has been
using Salesforce.com’s software to monitor the activities of its
salespeople and to track its entire inventory, as devices are
prescribed by doctors and dispensed to patients. Care Rehab
audited Salesforce.com’s security practices before agreeing to
PLATFORM for OPERATIONS


149
use the software. That audit included Salesforce.com staff
members showing Care Rehab how they secured the data that
was stored on their servers and reading documents describing
Salesforce.com’s security practices. “Their security is superior
to what we provide for ourselves,” said Barrett. “If you’re
Salesforce.com, you have to have the best people in security and
the best redundancies. [We] need to have the best salespeople.
I'm sure we aren’t the world’s best security people.”

Mexico Story
We’ll end this section with one personal story. A few years
ago, I had the opportunity to speak with the Secretary of
Treasury of Mexico. In my introductory remarks, I made a point
to tell the Secretary we could implement the Oracle On Demand
model either in an Oracle data center in Austin, Texas, or a data
center of his choice, presumably in Mexico City. I spoke for
about ten minutes and then asked him if he had any questions.
He said, “Tell me about your security.” I, of course, went
through all the security features from the data center, through
encrypted VPNs, through SAS 70 Level II audits. But I ended
my answer by sharing a conversation I’d had the previous week
with the CIO of OCBC Bank, one of the largest banks in
Singapore. In that situation, mid-way into my explaining the
breadth of on demand services the CIO said, “I want you to take
my HR and Payroll data.” I, of course, said yes, but asked him
CLOUD



150
why was he so adamant. He replied, “I know my DBAs can see
their salaries, their bosses’ salaries and their bosses’ bosses’
salaries. I know your folks can see the same information, but
what would it mean to them?”

It’s a perfect story to end with, since security professionals
agree that most IT-related attacks arise within the organization.
According to security professional Michael Bruck
26
, four-fifths
of all IT-related attacks arise within the organization. Bruck’s
estimate is consistent with a survey
27
of more than 200 IT
professionals, which found one-third of them admitting to
abusing their administrative access privileges to examine
sensitive data unrelated to their job functions. About one-fourth
reported knowledge of former employees who retained access to
sensitive networks long after their termination, with one-third
stating a belief that they could do likewise with little risk of
detection.

Customer Service
While many think of customer service management as defect
management, with any software that has a reasonable degree of
maturity, the challenge of customer service management is not

26
Bruck, Michael. Security Threats from Within. Entrepreneur.com, June 28, 2007.
27
Survey Reveals Scandal of Snooping IT Staff. Cyber-Ark Software Inc. Press release. May
30, 2007.
PLATFORM for OPERATIONS


151
around defects—it’s around delivering information, personal to
you. Just to drive the point home in 2004, Oracle’s Support
group received one hundred million requests for service. Now,
most people would think one hundred million requests would be
for software patches to fix defects. The analysis was quite the
opposite. Just one-tenth of one percent (0.1 percent) of the
requests actually resulted in any new software fixes; nine-nine
point nine percent (99.9 percent) could be answered by
knowledge in a library, existing patches that were contained in a
patch distribution system or in people’s heads—in short, known
information. So if you’ve ever been on hold with a call center, in
a post-Google era haven’t you wondered why was the
information so difficult to find?

Search-based Applications
Today's surface Web (which Google, Yahoo and others have
indexed) is thought to contain ten billion pages; or somewhere
on the order of 25-50 terabytes of information. But this surface
Web of less than 100 terabytes pales in comparison to the deep
Web of information contained in every business. How deep is
it? Consider that 1,030 petabytes (a petabyte is 1,000 terabytes)
of hard disk capacity was shipped in the fourth quarter of 2006.
The simple math says the deep Web is, at a minimum, 10,000
times the size of the surface Web. Sales, support, HR,
purchasing, and manufacturing information are all locked up in
CLOUD



152
relational databases, document management systems, and file
systems all around the world. While seeing or indexing the
information is one thing, the challenge is finding the relevant
information.

The opportunity to mine this deep Web has not been lost on
numerous companies. Google launched their Enterprise
Appliance in 2002 and, every year since, has produced a major
new release. Today they are deployed in over 9,000 sites around
the world. Independent companies like Fast, which was
acquired by Microsoft for over $1B, and Autonomy, the second
highest valued European software company after SAP, have both
focused on the challenges of enterprise search.

While these efforts have all been successful, enterprise search
in 2009 is today probably only a $500M business. This is small
when compared to the SQL database business, which today
probably exceeds $10B annually. So why the gap? Perhaps the
answer lies in tracing the history of SQL. Oracle was the most
successful at commercializing SQL and by 1987 was generating
$100M a year. But relational database technology has no
intrinsic business value (it’s just cool technology) and it wasn’t
until the advent of enterprise SQL-based applications like
Remedy, Peoplesoft, SAP, Siebel, and many more to make the
PLATFORM for OPERATIONS


153
technology useful to business. So if the pattern is to repeat
itself, we need to understand what it means to have search-based
applications. John Lervick, who founded FAST, said “Google
has been good and bad for search. Good because they made
search important; bad because they’ve made everyone think it’s
a search bar. Search can be so much more if you think of it as a
technology platform.” So what might these search-based
applications look like?

Udi Manber, who has been Chief Scientist at Yahoo, Chief
Algorithms Officer at Amazon.com, and was lured away a few
years ago to be VP of Engineering at Google, once described the
Internet in a very interesting way. He said searching the Internet
is like taking all the books out of the Library of Congress,
ripping all of the pages out, throwing them on the floor and
using a search bar to try and read them. If you think about it,
he’s basically right, and trying to read a book through a search
bar makes little sense. He recently unveiled a project called
Knol. Knol pages are “meant to be the first thing someone who
searches for this topic for the first time will want to read,”
according to Manber. The term knol means a “unit of
knowledge,” and has been seen by many as Google’s attempt to
compete with Wikipedia.

CLOUD



154
While Knol is targeted at the consumer Internet, a new
company, Openwater Networks, has engineered a search-based
application cloud service. Michael Rocha, President and CEO of
Openwater Networks, was previously the EVP for Oracle’s
$5B+ product support and services business. In that capacity, he
saw how the customers of complex applications struggled with
being able to understand their software, their business processes,
and their environments. While 99.9 percent of all of their
service requests were met with known information, it was the
cost and difficulty of delivering that information which inspired
him to bring modern Internet technology to the problem. The
search-based application fundamentally does three things. First,
it standardizes the language (much as SQL-based applications
have standardized processes) of a group, a team, or a business
unit. Second, the application is read and write. While this has
been a characteristic of all SQL-based apps few people think of
search-based applications as having this capability. Finally, the
search-based application acknowledges people add value to the
information, as opposed to traditional SQL-based applications,
which only control which people have access to which
information.

It will take a new generation of these kinds of search-based
applications to truly use the information buried in the Deep Web.
PLATFORM for OPERATIONS


155
But with greater and greater access to search technology,
whether in open source or in the cloud, there is bound to be
hundreds of search-based applications just as we have seen
hundreds of SQL-based apps.

Summary
Of all of the major functional areas, managing operations is
the most foreign to a traditional software company. But having
a system of repetitive, automatable operations is the key to low-
cost and high-quality customer experiences. This chapter has
highlighted some of the areas in the management of change,
problems, availability, security and performance. We’ve now
walked thru each layer of the cloud stack and now will focus the
remaining chapters on the important areas for anyone building
next generation application cloud services. How will you
market the application, sell the service, finance and organize
yourself for success?









156


MARKETING


So let’s assume you’ve built the world’s coolest application
and it’s going to revolutionize poultry management, or just the
next major release of your ERP application. How is anyone
going to know? Once upon a time you’d hire an “ad man” who
would come up with an ad campaign and then find out if you
could afford to run it in the Wall Street Journal, or just the Palo
Alto Weekly. You’d have to make sure your tagline was short
and sweet since you had no idea who would read it. As a result
many people think marketing means coming up with a clever
tagline and finding a way to make it into a Super Bowl ad. This
MARKETING


157
might have made sense when the production and distribution of
information was expensive, but with the advent of the Internet
and rich media, we clearly no longer need to think this way. In
this chapter we’ll focus on marketing your product. We’ll
discuss the fundamental economics, the traditional approaches
and some of the newer ways for you to educate the buyers.

Economics of Sales & Marketing
We’ll begin by considering the Webex case study by
becoming students of their P&L shown in Figure 7.1. The
funding line shows each round of funding they raised, which
totaled $107M prior to their IPO in 2002. You can also see that,
in 1996 and 1997, they operated at or near profitability. In these
early years, Min and Subrah financed their operations by having
a group of low cost engineers implement software and plowing
the profits back into funding WebEx (at the time called Active
Touch). This approach resulted in both founders maintaining
significant ownership of the company. You’ll notice the
explosive growth from 2000 to 2001. This was the by-product
of both a significant investment in sales and marketing and the
explosive growth of Web conferencing post September 11. You
should note that Webex’s spend in sales and marketing is
hovering around 35 percent, the single largest line item.
Consider the implication of this, given how we all view Web
CLOUD



158
conferencing as simple to use and simple to explain and simple
to buy.


Figure 7.1 WebEx P&L

Traditional Sales & Marketing
For those of you who have been doing high tech sales and
marketing this section will be old hat. Traditional marketing and
sales has broken the pipeline into a series of stages.

Figure 7.2 Traditional Sales & Marketing Pipeline
MARKETING


159

Early stages of the pipeline focus on getting awareness, then
leads. Leads are both nurtured and qualified ultimately
transitioning from a centralized marketing organization to the
field where the potential customer is engaged. Ultimately, the
big difference between consumer and enterprise sales and
marketing is that there is not one decision maker – enterprise
sales and marketing of imprecise, more expensive things is a
team sport. Consider Figure 7.3 and talk to anybody selling
business software. Even with a highly qualified lead the sales
person finds himself or herself in the roll of traffic cop (or
switchboard operator). Different constituencies within the
buying enterprise have questions that can only be answered by
members of the selling enterprise (or customers that have
already purchased – the reference customer). Scheduling the
phone calls, lunches, and web conferences is time consuming
and further elongates the sales process. No news here for any of
you who’ve sold enterprise technology.

CLOUD



160

Figure 7.3 Sales is a Team Sport

Hopefully thru CRM applications this process can be made
more efficient, repetitive and specialized. We’ll focus on a few
areas where marketing services have developed to improve sales
productivity.

Email Marketing
Email marketing, as an alternative to direct mail, has been an
area of considerable growth. This is largely because the
economics are so compelling. Emailing to a list of 1,000 people
costs $5 versus over $500 to send out direct mail. Companies
like Constant Contact, Lyris, Responsys have specialized in this
area. As a specialist they focus on the details. Everything from
the number of characters in the subject line (keep to less than 50)
to the time the email is sent out can determine the effectiveness.

MARKETING


161

Figure 7.4 Call at the right time

e-Loan Story
Scott Hilson, Director, Customer Services at Responsys, has
helped many companies implement successful email marketing
campaigns. As a simple case study, Siara Nazir, Director of
Online/Offline marketing at E-LOAN, implemented an email
program designed to remind and encourage prospective
borrowers to complete their mortgage applications. The
program involves integration with the E-LOAN Web site to
identify prospects that started but failed to complete the
mortgage application. It was executed in two phases. The first
phase, triggered 30 minutes after the application was abandoned,
reminded applicants to complete the process, provided a quick
link to the partially completed application, and encouraged
applicants to use phone support. The second phase, sent one
week later, called the applicant’s attention to the company’s
value proposition and also offered a quick link to the incomplete
application. Built into the program was the ability to perform
A/B testing on the messages sent to the customers. As a result
of this campaign, 32% more applications were completed by the
CLOUD



162
control group that abandoned but did get the automated follow
up email.

Search Marketing
A company’s Web site has become the face of the company
so improving the volume and quality of traffic to a Web site
from search engines via search results is clearly important.
While there is plenty of advice on how to spend money on ad
words, we all know we click on natural search results more often
than the ads on the side. By paying attention to how Web pages
are built and the content put in them, you can create natural
search results that drive traffic to your site without purchasing
advertising (and more importantly be significantly more
effective). Katy Roth, who after years of experience in
marketing Model Six companies formed her own consultancy,
has some fundamental advice for anyone trying to optimize
natural search results.
• Build a FAQ into your Web site to capitalize on potential
customers searching for reference or problem solving
information. This way, you become the expert on the
technology and the problem your company is trying to
solve. For example, Solarwinds, a network management
company, saw a lot of traffic coming to their site because
of their extensive network management content.
• Blogs are great for generating traffic. The content is
generally considered by search engines to be fresher and
more authentic, giving it higher weight in the results.
MARKETING


163
FiveRuns, a provider of monitoring and development
products for Ruby on Rails, developed content on their
blog that provided answers and guidance to Rails
developers. See the previous section on blogging.
• Have your community create content. iTaggit.com is a
good example of a company that reworked their forms to
pull in more community generated content. But they
didn’t stop there. They worked with their developers to
tune their page generation engine to produce better meta-
information that allowed their pages to come up to the
top of natural search results.
• Work with a search engine savvy developer to make sure
the meta content of your Web pages is optimized. There
is nothing worse than seeing a page with terrific content
with a generic browser title. Just check out Wikipedia for
some good examples.

Social Media Marketing
Facebook’s early success with college aged people has caused
some people to see social networks as just a tool to hook up or
find old friends. But Facebook has become much more. The
number of users has grown from 60M users in 2008 to over
300M with the fastest growing demographic being people 35-55.
Since Februrary 2009, Facebook leads Google, Yahoo and
MySpace in total time spent online by Americans (6% of total
time). Twitter has also enjoyed phenomenal growth growing
1,300% this year. Today more than 15 million users create 1,500
“Tweets” every second.

CLOUD



164
Perhaps the best way to think about Facebook and Twitter is
as a next generation telephone – and a very powerful telephone.
As an example consider the recent case where Facebook
simultaneously broadcast a soccer match in Spain for the
Spanish Facebook community. Within 30 minutes of the game
starting the video servers crashed. Why? Because once one
person commented and his social graph saw the comment (and
started watching the game) and another friend commented and
another, there were suddenly 100s of thousands of viewers.
Think of doing the same thing with a telephone and you’ll begin
to understand the potential of these social networks.

The application of these social networks to sales and
marketing is still in its infancy. Clara Shih attended the first
Facebook developer conference in 2007. With her friend Todd
Perry’s help, she developed Faceconnector which pulls
Facebook profile and friend information into salesforce.com.
Now instead of anonymous cold calling, sales reps could get to
know the person behind the name and title, and even ask for
warm introductions from mutual friends.

MARKETING


165
Clara’s counsel
28
for those thinking about using social
networking in your marketing efforts:
• Use hyper targeting as an opportunity to test new
audiences and see if your product can gain traction. Prior
to hyper targeting, it was often cost-prohibitive to do this
in any systematic way.
• Use the social graph and knowledge about relationships
across your audience to create a more personalized
online community.
• Think about using social networks to market yourself as
an individual. Your profile information, photos, and
friends all paint a picture of who you are and help to
establish your brand.

Get Notorious
This brings us to getting notorious. Once upon a time the
product marketing person interviewed the developers of the
product, spent weeks writing up a data sheet, deciding the colors,
quantity, and quality of the print that ultimately made it into a
folder at an event at the Hyatt in San Francisco. But there are
many more sources of information. You have internal experts
on the product in customer support, sales, R&D, product
management, field service, and consulting. Not only are your
people sources of information, but so too are your customers.
Customers of your product (the valued reference customers) that
you metered out onto a conference call with a carefully selected

28
The Facebook Era: Tap Online Social Networks to Build Better Products, Reach More People,
and Sell More Stuff, Claire Shih
CLOUD



166
sales person and prospect today can have their testimonials
videoed and made available on YouTube.

While the old world of low bandwidth, expensive
communication required that you needed to aggregate content in
the Internet behind the veil of “the company” today we should
be able to know what you think, what you believe, and what you
can contribute. As you develop your marketing programs, think
about whom in your company, in your customers, in your third
parties, should become notorious. Encourage them to write, or
speak, or do Webinars or videos and communicate their
expertise and more importantly their point of view. By bringing
more contributors into your marketing cloud, we will be able to
find the knowledge we need delivered the way we want it—in a
language and context that is accessible to each of us.

A few bloggers including Vinnie Mirchandani, Anshu
Sharma, Charlie Wood, and Ed Herrmann provide some advice
on getting notorious.
• Be notorious - Don't send a note from "your company.”
Reach out as an individual Name, title, and phone
number. Blogging is personal.
• Use video - but not an overproduced one. No talking
heads-show the product. Limit it to 30 seconds. Make
me think: “Wow, that's impressive!”
MARKETING


167
• Check out Digg - Look at the first few pages. See what
formats are popular, and use those. (Top 10's, video,
photo essays, controversy, "best ever’s, generally "this is
so cool!").

Educate & Select
Sales and marketing have been critical components of
traditional software companies and new software companies.
Perhaps the greatest change is, as we move closer to delivering
these applications as a cloud service, as they become more
specialized, and as the price point goes down, we must re-invent
the traditional account management model, the traditional
human network that has powered sales for the last twenty-five
years.


Figure 7.5 Selling Spectrum

As you begin to tackle this problem, consider the fact that,
last Christmas, very few people were sold a plasma TV. We’re
CLOUD



168
not saying there weren’t a lot bought, just that they were not
bought because of a sales person. Consider Figure 7.5. As a
technology industry, we have figured out how to sell expensive
items ($1M) that are not very precise (SAP Manufacturing) and
have grown large profitable businesses. We have also figured
out how to sell low price (<$100), precise items. You only have
to look at Amazon or eBay to see how successful this model has
been. Our challenge is how we effectively sell imprecise items
that cost $10,000.

We should take a lesson from the Plasma TV. The reason no
sales person sold a TV is that most consumers of Plasma TVs
did their research online; they figured out the right size screen,
read what others thought about the TV and, by the time they
were ready to purchase, had made their selection. Purchasing
became a transaction. Perhaps this is what can happen in the
future market for software. People will educate themselves on
the good and bad of the solution, talk to experts and other users
and, in the end, select the right answer for themselves. Selling
may be replaced by education and selection.

Summary
The Internet changes everything used to be a corporate
tagline. Perhaps it is no more obvious than in the art and science
MARKETING


169
of marketing. With a low cost means of delivering notorious
content, you now have an ability to deliver personalized
information to every member of the team that influences a
purchasing decision—when they want it, in the way they want it.

While communicating to people, whether they find you
through search, email, or social networks, you still need
marketing. In the end, no matter how simple we try to make the
software, the act of matching what a potential customer needs
and what you have to offer remains the role of a quality
marketing organization.









170

SALES


While SaaS or cloud computing might be all the rage, it
doesn’t take too long until someone says, “But this will
cannibalize our existing sales.” Sometimes it’s stated even more
clearly. Early in 2002, Ken Rudin, who today is the CEO of
LucidEra, was asked by Tom Siebel to lead the efforts for Siebel
On Demand. Rudin, of course, wasn’t sure if it was a good idea,
but Siebel finally convinced him. After about a year, Rudin was
able to work out the product strategy and business model. There
was a sense of urgency, as Salesforce.com was winning more
and more business. Finally, he was ready to tackle the sales and
distribution channel, and he scheduled a meeting with the head
SALES


171
of North American sales. The VP listened politely to Rudin’s
presentation and, at the conclusion, said, “Ken, that’s an
outstanding plan, but stay the f—k away from my accounts!”

This chapter is dedicated to sales and distribution. More than
with any other group, the solutions to the challenges here will
make or break any new business. In Models Four and Five, you
will need to figure out how to leverage your existing sales force.
In Model Six, the challenges will pertain to developing a new
sales force in a world where the economics don’t bring you
million-dollar deals that can fund the creation of a channel.

For the traditional software company, moving existing
applications to delivery as a cloud service with an existing
channel and product (Models Four and Model Five), your
primary advantage is that you already have a sales channel. But,
there are many challenges. Does it mean the comp plan will
change? Is this new on-demand group going to compete with
my sales force? How important is lead generation? In this
chapter, we’ll address some of these issues and also let you hear
from some successful sales and marketing executives.

Transforming Traditional Software Sales
Every Model One software company worries about
cannibalization when they consider a move to delivery of their
CLOUD



172
software as a cloud service. It doesn't have to be this way. If
you adopt Model Four, the existing sales force continues to sell
the software as it always has. Whether that’s a perpetual or term
license, nothing changes. The only change is they have a new
delivery option (one which they can be compensated for). Do
you want to manage our software or do you want us to manage
our software? Do you want to capitalize and pay for the
software, hardware, third-party software, etc., up front or
expense it over time?

Joe King, who led sales for Oracle On Demand for seven
years and today is Group Vice-President of JDA’s on demand
business, has first-hand experience in how to sell when you’re in
Model Four. His first piece of advice is to make sure you find a
team of sales professionals that have a mix of experience in
selling both traditional software products and services. Next,
organize the sales teams around current delivery capabilities.
Ensure that the sales team is setting proper expectations with the
customers and selling a solution that your company has the
ability to deliver on today. Unlike selling traditional software,
Joe says, “Don’t sell futures!” Finally, train the sales reps to
find customer challenges that can be solved by having you
provide the complete solution without letting the customer
dissect the solution into its components. In other words, it’s
SALES


173
important to discuss the underlying infrastructure that supports
the solution, but remember the customer is not buying servers
and data centers separately from your complete solution.

And when you get to the CIO or IT director, who will
inevitably say: “I can do this” (because they have), you need to
be armed with the evidence that when a software company
delivers the service in a repetitive standardized way, they will
ultimately be better and cheaper at insuring security, managing
an upgrade, or identifying performance problems than you as a
single isolated island having to figure it out for yourself.

Selling New Application Cloud Services
For those of you building a Model Six business selling
software as a service is in many ways no different than selling
traditional software in Model One. There still needs to be value
in your software in solving a particular business problem. What
is different is the economics of the sales process. In Model Six,
the average selling price is going to be significantly under
$100,000 (for the first year), so no longer can you fund the
software sales by the $1M up-front license fee. As a result, the
traditional sales person who manages an account and makes a
good living sitting in a few accounts is unaffordable in the new
world.

CLOUD



174
While that aspect is worse than our old world, there are some
things that are better. Since the software is either “always
deployed” or “ready to be used,” or “quickly configured,” the
idea of scheduling custom demos is a thing of the past. Now
with adequate Web marketing and standard demo deployment,
some group of customers can try your product and then proceed
to use it in small work groups.

Joe Graves, CIO at Stratus Computers, a $250M a year
computer company, is a big user of application cloud services.
Today he uses Salesforce.com, QuickArrow, Eloqua, Xactly,
SpringCM, Ceridian, Halogen, WebEx, and Postini. Joe relayed
a story of how different traditional software companies are from
new Model Six companies. A Model One software company
offered to do a free six-week discovery process and scheduled a
demo for eight weeks later. Stratus went with a Model Six
solution and deployed the whole system in seven weeks—
worldwide.

Salesforce.com and WebEx both can attribute their high
growth and early success to having products that could be used
simply and economically by small groups. Remember, the good
news is Model Six products are reasonably precise versus the
large application footprint of traditional enterprise software. The
SALES


175
bad news is that the average selling price is considerably lower
in Model Six than in Model Four. Salesforce.com averages
about 20 seats per customer, with an average price per seat of
around $70 per month, meaning that their average yearly
contract is under $20,000. WebEx’s high-end customers
average around $25,000 per year with 50 percent of their
customers having an average yearly contract of less than $1,000
per year. The point of both of these numbers is that an average
deal size is nowhere near the average selling price of traditional
software. A sales person can no longer depend on landing a few
big whales to make quota.

So from a company perspective you need volume. And when
you need volume, you need a sales machine. WebEx’s approach
has been to spend nearly $100,000 in marketing per sales rep per
year. This includes free trials, daily demos (scheduled and
executed from India), Webinars, and email campaigns. Every
lead from this marketing is categorized into three buckets. High
priority leads are responded to within 24 hours; the next level
within 72 hours. With an eight-to-one conversion ratio and a
fifteen-day sales cycle, you can see they have perfected a sales
engine.

Salesforce.com has also had a few tricks up their sleeve.
Aaron Ross was Director of Corporate Sales with
CLOUD



176
Salesforce.com from 2001 to 2006. We asked him what advice
he would have for establishing high velocity sales. He started by
saying something that might seem counter-intuitive: “The size of
your sales team does not drive growth.” In the past, a VP of
Sales would drive growth by hiring more sales people, who
would work their patches to generate leads and close more
business. Today, that doesn't work. In Ross’ opinion, hiring
more sales people is not the main engine of growth. Instead,
generating more leads, which sales people close, is what drives
growth. Likewise, instead of obsessing over the number of sales
people in your team, obsess even more over your sales
productivity. There just are not enough great sales reps to hire
out there, so helping your current team do more with less is as
important as hiring.

Ross went on to say: “Give up Control.” Historically, the
sales organization wanted “control over the sale.” This was
possible, because it was very challenging for customers to get
unbiased information about vendors. The Internet changes
everything. Today, prospects can find out as much as they want
about you before they ever call you. Vendors can’t control the
flow of information. The best strategy is to embrace this, and
instead, think about how much information and control over the
SALES


177
sale you can put into the hands of the prospect. How much can
you educate the decision makers and influencers?

Ross’ next piece of advice again might sound counter-
intuitive to old-school sales people: “Stop the ‘Always be
closing’ mindset.” Buyers have seen all the sales tricks and
programs, and anything artificial irritates them. So instead of
“Always be closing,” focus on building a sales team that is
constantly asking, “Are we a match?” You can't afford to waste
much time (money) on prospects that aren’t a great fit.

Finally, lead generation and lead qualification are as
important as closing. Most VPs of Sales focus too much on the
second half of the sales process: sales deals in the pipeline and
what can close. They tend to ignore the first half of the process:
lead generation and lead qualification. Yet, the more investment
sales put into both the quantity and quality of leads, the higher
the quality and the larger the sales pipeline will be...making sales
results easier. Here are two specific examples of common
mistakes: not hiring enough junior sales people to qualify
inbound leads, and not creating a dedicated team that does
nothing but outbound prospecting. One way to fail at generating
leads is to expect your own quota-carrying sales reps to do it;
they’re terrible at it, they hate doing it, and as soon as they get
some pipeline, they become too busy to prospect anyway.
CLOUD



178

Contracts
Contracts vary significantly from model to model. Since the
earliest forms of selling software as a service were large
outsourcing deals (Model Three), many contract issues for
Models Four through Six carry with them some of this legacy.

The outsourcing (Model Three) business is still a growing
business around the world. In Q1 2007, Forrester reported
companies based in Europe, the Middle East, and Africa
(EMEA) signed 75 large outsourcing deals in the quarter—with
a total deal value of almost €5.7B. If you were to look at the
contracts for those deals, you might discover the same thing that
Bechtel’s CIO, Hank Leingang, found when he plowed through
its contract to outsource desktop support and network
management to Electronic Data Systems: “It was three inches
thick.”

With this degree of complexity, there is a cottage industry in
providing advice to companies negotiating contracts. TPI is a
leading global outsourcing advisory firm. They published a
paper providing advice to customers renegotiating outsourcing
contracts. When asked about the duration of renegotiation, 85
percent of those surveyed indicated the restructuring process
SALES


179
required less than a year to complete. Fifty-five percent finished
renegotiations in less than six months, and 13 percent finished in
less than three months.
29


At the other end of the contract spectrum is the online click
wrap agreements. For examples, take a look at the Yahoo terms
of service or the WebEx contracts. Typical Model Six terms
allow customers to pay on a month-to-month basis, so contracts
can be terminated with 30-days notice; other vendors require a
two-year commitment. Furthermore, some vendors choose to
enforce a penalty against the total value of the contract if a
customer wants to cancel the deal prematurely.

Service Level Agreements
In almost all of the contracts from Models Three through Six,
the subject of SLAs (Service Level Agreements) will come up.
In Model Three, contracts like these SLAs are heavily
negotiated. If you want to spend a couple of frustrating days,
you can lock five experts in a room and try to define what
uptime means and what penalties you should pay if the uptime is
not met. Binding SLAs to contracts can vary. Companies like
Salesforce.com offer SLAs on contractual terms on a case-by-
case basis. In other cases, such as Blackbaud On Demand, the

29
Restructuring Outsourcing Agreements: An Indication of Failure, or a Tool to Increase Value?
TPI Report, January 2007.
CLOUD



180
contracts stipulate 99.9 percent uptime and are a standard part of
the contract. There is a 99.9 percent uptime guarantee and a
commitment on the specifics for data handling and server
maintenance. “We occasionally modify the terms slightly for
customer’s specific needs, but it is a competitive agreement,”
said Blackbaud’s CEO Marc Chardon. “We're very strict on
ourselves, as this agreement governs what we do operationally.”

Taleo provides penalties in their contract if they don’t meet
certain uptime standards. Although the financial penalty might
seem minimal to the individual customer, they point out, if they
are writing a check to that customer, chances are they are writing
a lot of other checks. So the total financial incentive for Taleo to
keep the system up and running all day and every day is
enormous.

In either case, the important point is that a user of an
application wants the service to perform, so you should first
focus on what service level your operations and development
group are engineering to deliver. After that, you can decide to
what extent these metrics are embedded in a contract. As noted
earlier, Oracle pioneered a different approach, simply called the
“Nordstrom’s guarantee.” Service levels in availability,
security, problem, change, and performance management were
SALES


181
published and reported. However, if the customer was not
satisfied with the service for any reason, they could request a
rebate of 20 percent each month, no questions asked. In four
years of offering the Nordstrom’s guarantee, less than a handful
of rebates were paid out.

The Nordstrom’s guarantee was an effort to separate the issue
of penalties from a description of the service level metrics and
objectives. Writing these down and measuring them is key to
improving the overall quality of the service. Let’s highlight the
major areas you should focus on.

• Change Management – Define how often upgrades to
new releases will occur. Will you move all customers at
once, or will you stage them over a three-month period?
• Availability Management – Many contracts call out
percentages of availability (99.5 percent, 99.7 percent,
99.9 percent, and 99.99 percent… They all sound about
the same, don’t they?). This begs the question of
whether this refers to unplanned or planned outages.
You should consider defining availability by total
number of outage minutes per month, per quarter, per
year, and do that for planned and unplanned outages. It
will cause you to focus on how rapidly you recover,
which is a much more important metric.
• Security Management – Have you completed a SAS 70
audit? How frequently are backups taken? Where are
backups stored? How fast can data be restored? Under
normal circumstances or in the case where there is
complete site failure because a disaster has occurred.
CLOUD



182
• Customer service management – When a customer has a
question about your software (these could be defects or
just usage), what level of service will you be providing?
• Performance Management – What response time should
customers expect? How much disk space is included?
How quickly should reports run? This will be a difficult
area, but providing some reference implementation to
benchmark on a release-to-release basis will help you
provide performant apps. For a service level that is
reported on every transaction, just look at the upper
right-hand side of your Google search results when you
search for “SaaS”.

You would be wise to separate the definitions of your service
levels from the legal and contractual side. Become a student of
WebEx’s and Salesforce.com’s contracts and terms. Some will
push you towards EDS or IBM Global Service agreements.
Instead, be proactive about the service levels you will provide
and publish them to your customers. In the end, remember the
customer wants the service to work.

Sales Compensation
Marc Chardon, CEO of Blackbaud, points out the universal
truth in effective sales compensation: You’re looking for sales
people to respond to the metrics that you give them. “We try as
much as possible to make the sales rep indifferent to selling an
on-demand or on-premise version,” he said. “If you create a
level playing field in the sales compensation, and if the solutions
SALES


183
are roughly as easy to sell, sales reps should be, and usually are,
good at evaluating the shortest time necessary to get their
commission.”

When you look for expertise in sales compensation you’d
have to consider the team at Callidus Software. They have some
guidelines that are applicable across a number of our seven
business models.
• Keep your compensation plans straightforward. Many
companies come up with plans with dozens of variables
and multiple levels of intricate dependencies.
• Make sure you have the data—and that it’s clean.
• Report regularly, transparently, and consistently with the
plan type and organizational style. One good
(transparent) report delivered to payees at the right
frequency keeps the focus and motivation where it
should be.
• Credit directly for sales compensation. The most
effective plans credit the primary seller along with those
directly connected to that seller in the chain of command.
• Create a transparent dispute resolution process. Even if
you are able to follow every piece of advice above, errors
and problems will happen.
• Analyze and change on a predictable schedule. The
smartest and most successful companies are constantly
evaluating their sales performance data and are not afraid
to make changes to compensation plans.

CLOUD



184
Installed Base Sales
Most of the focus on selling has been on getting the new
customer, again a by-product of selling a product for large
upfront money versus driving revenue on a monthly or yearly
basis. While traditional sales has been on trying to know
something about a customer you don’t have – for the first time
companies can now focus on learning more about the customer
they do have and based on that information personalize the
offering using that knowledge.

Nowhere is this more obvious than in the case of Amazon.
But how can this be applied to higher ASP, less precise
products? Here we can take a lesson out of the notebook of
computer architecture. Computer architects have struggled with
how to make computers faster and faster. While turning up the
clock speed is one answer, another answer lies in being more
efficient about how an instruction is executed. Below is a
diagram showing the typical execution of an instruction. Years
ago every instruction took 5 clock cycles to execute, as a result
seven instructions would take 35 ticks of the clock. But over 10
years ago some smart guys figured out if you could separate
instruction execution into 5 stages, 5 parts, each taking the same
amount of time (1 tick of the clock) then instead of 35 ticks to
get 7 instructions executed you could do it in 11 ticks. So with
SALES


185
the same clock speed the computer was in the limit five times
faster.


Figure 8.1 Pipelining

So what does this have to do with sales? Our marketing &
sales pipeline is no different only the clock ticks are not so fast.
The conventional way of managing the pipeline is no different
than the traditional way of instruction execution – we finish one
deal, one instruction, before we start the next. But indeed the
same principles can be applied, and were applied to sell a
standardized Model Four service to 10,000 companies who
already owned Oracle applications. The trick is to engineer each
stage of the pipeline to last a fixed period of time (e.g., 2 weeks)
and then move customers from one stage of the pipeline to the
next. Then each stage could be optimized and specialized and
over time sales transactions could be much more parallel than
they are today. Anyone selling to a large installed base so
reconsider how – and not just treat it the same as potential
CLOUD



186
customers – where you have no idea who they area, nor what
experience they’ve had with the product.

Indirect Channels
The last major generation of computing was called client-
server computing. Its rise was not only driven by the availability
of low cost hardware (Dell, Compaq, etc.) and low cost software
(Microsoft, Corel, etc.) but also by the creation of a new channel
of distribution. Names like Ingram Micro, Tech Data and CDW,
today all multi-billion dollar companies, provided a low cost
indirect channel that matched the price of the technology
products.

This client-server indirect channel has been built on
delivering “boxes and bodies” to implement and deliver the
application. Unfortunately, whether you’re in Model Four,
Model Five or Model Six, you are now in competition with these
traditional channel players. In early 2006, I had the opportunity
to spend a day with one of Microsoft’s largest resellers for its
Great Plains Software. While the reseller saw the value of a
Model Six business, they were as addicted to the “perpetual
license drug” as the vendor. Allow me to explain. Today when
they sell a $200,000 license, they get 35 percent of the sales
price, or $70,000. With $70,000, they can afford to pay the sales
SALES


187
rep to drive the sales cycle and then perhaps make more money
by doing the implementation of the software to the specific and
unique requirements of the customer. Consider the same
scenario in subscription view. Now the $200,000 might be
spread over four years. That’s $50,000 per year. Thirty-five
percent of that number is now only $17,500—and the
implementations are far more standard.

Indirect channels have been important to the software
business, so it’s no wonder people are trying to figure out what a
next-generation channel will look like. Erin TenWolde and
Darren Bibby of IDC published a research paper
30
in late 2007.
Key highlights from their study were:

• There are still nascent partnering business models in the
SaaS ecosystem. Based on interviews with SaaS
providers, there is still a lot to learn and most of the
activity is occurring on a trial-and-error basis.
• Traditional channel partners, particularly value-added
resellers (VARs), have some potentially big challenges
ahead, especially in terms of securing top-line revenue in
resale scenarios. This will have implications on the
valuations of VAR businesses, which traditionally have
been valued based on a multiple of revenue.
• With many traditional IT challenges removed with SaaS,
IDC believes those new solutions providers will emerge
that are more business process oriented rather than

30
TenWolde, Erin and Bibby, Darren. The Emerging SaaS Channel. IDC Report #208547,
September 2007.
CLOUD



188
technology focused. Those partners that can develop
discrete expertise in niche areas will be well positioned
to take advantage of the vast SaaS opportunity.

Axel Schultze, who founded Computer 2000, the largest
reseller of PCs in Europe, also founded a Model Six company
called Blueroads. He’s uniquely positioned to discuss what a
next-generation channel might look like. Schultze remembers
1981, when IBM announced the PC, Compaq was founded, and
Microsoft introduced BASIC. He saw the first three to five
years of the PC industry focused on replacing existing midrange
computers like IBM /34 or DEC PDP 11 with personal
computers. This was a tough strategy and a cumbersome fight
against established computer manufacturers, established
software, established consulting processes and, most
importantly, established sales channels.

Can we learn from the PC industry development in the 80’s?
What happened back then that led to the explosion of the PC
industry and the enormous rise of companies like Compaq, IBM,
Microsoft, Lotus, Ashton-Tate, and Intel? An interesting, almost
unnoticed change happened: young startups entered the market
and didn’t sell into the established businesses, but into a new
market segment—companies below $50M in revenue. They
started selling in companies who never even wanted a large
SALES


189
midrange or mainframe computer. And along those lines, a
channel emerged that helped sell and install those PCs and PC
software for $1000—a price a mid-size systems integrator
wouldn’t be bothered with. Axel boiled it down to a simple
premise: “There is interesting software for a large variety of
customers and industries. What is missing are the new channel
companies.” Will history repeat itself?










190

FINANCE


You might have a killer application and a market that is
rapidly growing, but if you don’t have money (and people) it
will be hard to turn it into a reality. This chapter focuses on the
money – some of the key metrics that you’ll need to track in
order to insure a long running profitable business. We’ll discuss
monthly recurring revenue (MRR), revenue recognition, churn,
the Magic Number and how much it takes to launch a new
business.
FINANCE


191

Revenue Recognition
Many traditional software companies are concerned about
moving to Models Four through Six because of something called
‘revenue recognition’. If a Model One company sells software
for $1M, they recognize the million dollars on the day they sell it
and it goes into the revenue for that quarter. If they were to sell
that same system as software as a service, they may get $20,000
per month for the next 10 years. The same dollar amount is
spent on applications, but logged in accounting books differently
under Model Six. As a result, traditional (Model One) software
companies have been concerned about experiencing a negative
short-term impact on license revenue. This, of course, assumes
that they choose to move to Model Five with their existing
license business versus working in Model Four.

In the traditional software business, the largest part of
revenues stems from vendors’ license fees associated with
software. A Model One company recognizes revenue from
license fees when the software is shipped to the customer.
Traditional software revenue accounting requires that the
vendor’s fee be allocated to the various elements based on
something called vendor-specific objective evidence (VSOE).
VSOE is limited to the price charged by the vendor for each
element when it is sold separately. This requires the deferral of
CLOUD



192
revenue until VSOE can be established for all elements in the
arrangement or until all elements have been delivered. If
customer support is the only undelivered element in the
arrangement, however, the entire fee can be recognized ratably
over the term of the customer support contract. In addition,
recognition of revenue must be deferred if undelivered elements
are essential to the functionality of any delivered elements.

Priscilla Morgan, former head of finance and business
operations for Oracle On Demand says, “For those pursuing
Model Four, the simplest way to think about providing software
on demand is to treat the revenue the same as you have treated
customer support.” In fact, some of the same contractual terms
and sales methodologies can be used. Contracts are yearly,
subject to renewal by your support renewals teams. You’ll also
find out that your services will be much more specific than the
terms you already have for product support. This way, one-time
charges, such as implementations or customizations, can be
treated the same way you have managed professional services
contracts. In these cases, you’ll have to recognize revenue based
on milestones and the level of work that has been completed,
but, once again, this should be no different than how this is done
in your traditional software business.

FINANCE


193
Monthly Recurring Revenue (MRR)
Tien Tzuo, CEO of Zuora, points out that you’re going to
need the right metrics to run any Model Six business. A
measure used by many Model Six companies is MRR (Monthly
Recurring Revenue), which by the way, should also be used by
Model One companies to track their support revenues.

MRR is a powerful idea but as such has both upside and
downside. On the upside assume your company gets $50k of
new MRR during your first twelve months of revenue
generation. The rule of 78 says the company will generate a total
of $3.9 million ($50k * 78) of revenue over that year. [Note: 78
is simply the total of the revenue months during the year;
12+11+10+...+1]. Furthermore – next year if it sells no new
business revenue the company will still grow to $7.2M in year 2,
since in the last month the company was doing $600K in MRR.

But what happens if the company does only $25k of MRR
bookings. Now the first year revenue is only $1.95 million. And
the company just missed its revenue number by $2 million.
Worse yet, if your fixed operating costs are $600k per month,
then the business is close to covering fixed operating expenses
under the $50k MRR bookings scenario, but now 12 months
away under the $25k MRR bookings scenario; assuming no
CLOUD



194
churn. Meaning you’ll need another $2M on cash/investment.
Which of course brings us to the subject of cash.

Cash
As Subrah Iyer commented, Traditional Model One
businesses have been great since with typical software license
being sold for an Average Selling Price of say $500,000 could
bring you the cash you needed to hire more people or buy more
computers. Unfortunately, this is not exactly the case in a lower
ASP, Model Six business model. However, if you could get a
customer to change their payment terms and pay for the service a
year in advance or better yet three years in advance it would do a
lot to help your cash needs.


Figure 9.1 Power of Payment Terms

Cost of Sales: When to Hire for More Sales
As we’ve repeated Model Six companies don’t have the
advantage of Model One companies, which historically have
FINANCE


195
derived a large up front, license fee when a sale is completed.
Moreover, the service requires significant infrastructure
investment. “Every time we add an incremental customer, it
costs us more money that quarter—it costs us more cash that
quarter” explained Josh James, CEO of Omniture. To get a
sense of the cost of service, Omniture operates a 15,000-server
cloud for its 5,000 customers, and processes almost a trillion
transactions per quarter. The financial consequences look
exceptionally dire when expressed according to GAAP
accounting rules, which force the cost to be expensed upfront.
“When we were really stepping on the gas in 2004 we were
GAAP unprofitable,” said Josh. In 2005 Omniture’s negative
free cash flow was in excess of $22M—more than its total
annual revenue—and it had spiked higher at some points. “It is
a scary moment. Even when management understands the game
plan and has the full support of its financial backers, it takes
courage to plow ahead regardless,” he said.

But when is it the time to step on the gas? When is it the time
to ramp your sales force? Josh and Omniture have been relying
on a simple calculation, shown is Figure 9.2 that produces what
they call a Magic Number.

CLOUD



196

Figure 9.2 Magic Number Calculations

Figure 9.3 shows the analysis for a group of companies.
According to the analysis any time the Magic Number is .75 or
more, invest in more sales people and if the Magic Number is
less than .5, there’s probably something wrong with your
business.

Figure 9.3 Magic Numbers for Model Six Companies

Churn: Cost of Keeping a Customer
Another critical metric for recurring businesses is churn, or its
opposite, renewal. But there are many ways to calculate churn.
For example, for any given period, what percent of the business
FINANCE


197
at the start of the period is still there at the end of the period?
The part that is lost is considered the churn. For companies with
longer-term contracts, a renewal metric may be more
appropriate. For a given period, such as a month, what percent
of the business that is up for renewal actually renews?

Financial Systems
Many companies dramatically underestimate what it takes to
run a Model Six business. In a Model One software company--
you express interest in purchasing my products, I write up an
order form, ship you the products, send you an invoice, collect
payment, and the transaction is completed.

Not so in the recurring revenue world. The first order is only
the start of the relationship. Each month, I can measure how
much of my service you are using and compute your bill. In
addition, at any time, the customer may choose to change the
service. Perhaps you have purchased a cell phone plan, and now
you want to add text messaging. Perhaps you started off with 10
licenses of an application, and now want to add another 2
licenses. Or perhaps you subscribe to delivery of the New York
Times, and you want to suspend delivery for 2 weeks while you
are on vacation. Subscriptions are a living, breathing
representation of the relationship between you and your clients,
and every change to the subscription must be handled by all your
CLOUD



198
back end systems from quoting, to billing, to order provisioning,
and through to collections.

For any Model Six company, having the right billing and
payment systems is important to growing and scaling the
business. Any billing system, whether you buy or build must
give you pricing and packaging flexibility. At Salesforce.com,
Tien Tzuo, today CEO of Zuora, learned early on the importance
of having the right pricing & packaging strategy in order to grow
a subscription business. “When we first started, we thought the
right strategy was to keep our pricing simple. We priced our
sales force automation (SFA) service at $50 per user per month,
and we thought that would be the price forever.

The market, though, had other ideas. Many tiny companies
loved the idea of an on-demand service to manage their sales
force, but they told us that $50 a person was too steep a price.
At the other end of the spectrum, large companies like Autodesk
told us they actually wanted to pay more -- but in return, they
wanted more features, and stronger customer service.

We quickly realized that a one-size-fits-all pricing model
would never work. Different companies had different needs, and
different price points. That's when we embarked on our
FINANCE


199
packaging strategy that ultimately led to a Professional Edition
at $65 per person per month, an Enterprise Edition at $125 per
person per month, and a Group Edition at $995 per year for 5
users.”

If you look at other subscription companies like Netflix and
Zipcar, you see the same evolution in their pricing models.
Netflix started off in 2000 with a simple model -- $19.95/month.
Fast-forward almost a decade, and Netflix now has over 9 plans
hitting multiple price points. Their most popular plan is
$16.99/month, labeled as the 3-DVD-at-a-time plan. But
subscribers new to Netflix can dip their toe in the water with the
$4.99/month 1-DVD-at-a-time plan, and heavy users can
upgrade all the way to the 8-DVD-at-a-time plan for a whopping
$47.99/month

Similar, Zipcar started off with a $50/year membership fee
for their popular car sharing service. Today, Zipcar also has
renamed the original plan as the "Occasional Driving Plan", and
introduced 4 new price plans called "Extra Value Plans" for its
heavy users. The basic lesson to learn here is that different
customers have different needs. Customers want choice as to
how they consume your services -- and how they pay – and the
financial systems for service businesses need to reflect this
degree of flexibility.
CLOUD



200

Financial Analysts
As a new company, you’re going to hope that, one day; the
financial analyst community will be talking about you. So what
are they going to be interested in? How do they evaluate
investments for the public markets?

Laura Lederman, a William Blair & Co. financial analyst who
focuses on new software companies, sees three important
evaluation criteria. The first thing she looks at is total available
market. “It’s the most important factor. The larger the market,
the better the investment is likely to be.” Number two on her list
is the quality of management team. From her perspective, it’s
“one of the reasons Salesforce.com, Vocus, and Concur have all
done so much better than other players in the market.” And,
finally, her third criterion is to evaluate the level of competition:
the more competitive the market, typically the more difficult the
investment.

Venture Capital
If you were a student of Salesforce.com and you’d know that
it took $60M+ in capital to get the company to an IPO. This is
also the same type of number in Webex’s case. Of course the
money was well spent since particularly in the later stage
FINANCE


201
investments there was a direct correlation between investment in
sales and revenue. Again, like WebEx, CRM took about six
years to reach $100M in revenue. Students of traditional
software companies know most of them struggle to get to $50M,
and on a much slower growth curve. It is no wonder that both
WebEx and Salesforce.com have high valuations relative to sales
and earnings.


Figure 9.4 Salesforce.com P&L

Some of you reading this are starting up new companies and
contemplating being “like Mark”. Ann Winblad began her
career as a systems programmer, and in 1976, she co-founded
Open Systems, Inc., an accounting software company, with a
$500 investment. In 1998, she co-founded Hummer Winblad as
the first venture capital firm to invest exclusively in software.
Their first investment in a Model Six company was Employease
in 1998. In a recent lecture at Stanford, Winblad discussed her
CLOUD



202
view of software as a service, and identified a number of
advantages. Having forward predictability, being independent of
capital or IT budgets, being able to address unserved markets,
and always “serving” the customer are all characteristics
Winblad sees as being better.

Her key characteristics of a Model Six company are:
• Can be sold and delivered over the Web (via browser)
• Single instance, multi-tenant architecture
• Can be sold to small and medium size businesses (SMBs)
as well as enterprises
• Requires little customization
• Requires low professional services
• Pricing model is subscription based (by time period, user
or transactions)

Summary
Finance and business operations for a software product being
delivered as an on-demand service are very different than what
we’re used to in the traditional software model. But if the sales,
operations, and R&D can come together under a unified business
model, the growth rates and valuations can be astonishing.










203

HUMAN RESOURCES


Software is a people business. Whether it’s cloud computing,
Web services, software as a service, or software on demand it
makes no difference. In this chapter, we’ll discuss some of the
people challenges: finding or teaching the right skills,
establishing the right organizations, and insuring you have the
right people in the right jobs. If you’re an existing software
company, you’ll face the Innovator’s Dilemma. The challenge
of growing a new business is difficult and getting the right team
and the right model will be key. If you’re a new software
company, finding the right people and organizing them
CLOUD



204
appropriately will be a constant challenge. But, all of this begins
with people.

Economics
If we again use Webex as a case study of a successful
company you can see in Figure 10.1 how the company’s growth
was tied to it’s ability not only to raise money, but also to hire
people, and lots of people.


Figure 10.1 Hiring ramp at Webex

And by the way, not only to hire people, but also to fire
people, as you can see in the year 2001. As a result much of the
HR application cloud services have focused around recruiting
(Taleo, Kenexa, etc.) and performance management
(Successfactors, etc.).

Right People, Right Stage
For many of us, Geoffrey Moore’s landmark book, Crossing
the Chasm, best describes the lifecycle of a high tech company.
In this and in subsequent books, Moore describes the seven
major stages of a successful venture:
HUMAN RESOURCES


205

• Early Market – A time of great excitement when
customers are technology enthusiasts and visionaries
looking to be first to get on board with the new paradigm
• Chasm – A time of great despair, when the early-
market’s interest wanes but the mainstream market is still
not comfortable with the immaturity of the solutions
available
• Bowling Alley – a period of niche-based adoption in
advance of the general marketplace, driven by
compelling customer needs and the willingness of
vendors to craft niche-specific whole products
• Tornado – a period of mass-market adoption, when the
general marketplace switches over to the new
infrastructure paradigm
• Main Street – a period of after-market development,
when the base infrastructure has been deployed and the
goal becomes to flesh out its potential
• Assimilation – the technology loses its discrete identity,
moves into decline, and is supplanted by a new
technology paradigm

While successful companies need to go through these stages,
it’s important to realize that people need to go through these
stages as well. We often see that the CEO of a company at the
early market stage is not the CEO of the company when it
reaches Main Street. Of course, there are the rare exceptions of
the individuals with the gift of being able to adapt, learn and
change as they cross through each of these stages. Names like
Jobs, Ellison, and Gates are the clear exceptions to the rule.
CLOUD



206

But for the rest of us, what are the characteristics that are
most important in each of the stages? Tom Kippola, co-author
of The Gorilla Game, recently said, “As a company is trying to
cross the chasm, the title CEO does not mean Chief Executive
Officer, instead it should mean Chief Experiment Officer.”
While, in hindsight, it might look like some companies took a
straight line from inception to Fortune 500, the reality is, there
were many left-turns, right-turns and U-turns along the way.
Without the ability to quickly run experiments, learn from them,
and then press on, most would have failed early in their
lifecycle. Of course, there does come the day that
experimentation and cheerleading need to be balanced with
operational excellence. Making the quarterly sales numbers,
delivering releases on time, and sticking to a budget are all
requirements for any successful business. The mistake
companies often make is putting an operationally excellent
leader in place before they’ve crossed the chasm, or forgetting to
get their operational house in order as they’re headed down to
Main Street. Getting the right people at the right stage, that’s the
ticket.

HUMAN RESOURCES


207
Right People
As with any team, having a good coach and quarterback is
essential, but having the right people at all of the skill positions
is the key to success. Delivering cloud services requires
operational people, development people that are different from
what we’ve seen in the traditional software world.

As Evan Goldberg pointed out, your engineering group needs
to both have a “high tolerance for change” as well as be
“accurate and careful programmers.” Furthermore, you’re going
to need what he calls next generation DBAs, operations people
who understand both software development and software
delivery. Lynn Reedy recommends you have the VP of R&D
work for the VP of Operations; or vice versa. By doing so, it
forces conflicts between the race for new features and the ability
to manage change in a reliable way. Clearly, this is something
we never had to do in traditional software companies.

So, where are you going to find these skilled people? One
place to look is in the traditional software product support
organizations. People who have had to optimize performance or
availability of software implementations in the field will have
been exposed to some of the key processes and challenges
involved with delivering Web applications. Often times, these
individuals are leading premium support services and are located
CLOUD



208
at or near a customer site. The challenge for these people is that
their tendency will be to customize a solution for each customer
installation. You’ll have to find the people that can generalize
the lessons learned in the field and apply them in a standard,
uniform way.

You may also discover some of the greatest expertise to be
individuals who have been in end-user operations where they
have had to manage applications for a large internal population
(e.g., the people who ran American Airlines SABRE). Many of
these people have run large-scale operations. Rick Dalzell, who
recently retired from Amazon and was their Senior Vice
President of Operations and CIO, was recruited out of Wal-Mart.

Culture
In 1998, 24-year-old Tony Hsieh sold his company, Internet
advertiser LinkExchange, to Microsoft for $265M. What some
people don’t know is he sold it because he stopped wanting to
come into work. While LinkExchange had been a successful
company, he never focused on the company culture, the common
values shared by everyone in a group. So it should be no
surprise that when he invested in and later became CEO of
Zappos, the leading online shoe retailer he made culture a
cornerstone of the company. Culture starts from the hiring
HUMAN RESOURCES


209
process, where they ask, “Which superhero would you want to
be?” and continues thru their four weeks of training where every
employee (even VPs) answers phones in the call center and ships
shoes out of their distribution facility in Kentucky.

After the training, they are offered $2,000 to leave the
company—no questions asked. This "quit now" bonus, which
started at $100, is designed to ensure employees are there for the
right reasons. In 2008 less than 1% of the trainees accept what
the company calls "the offer."

Zappos.com also publishes a "Culture Book" every year. It’s
several hundred pages of employees from all across the company
describing what the company culture means to them. The
company itself maintains 10 core values:
1. Deliver WOW through service
2. Embrace and drive change
3. Create fun and a little weirdness
4. Be adventurous, creative, and open-minded
5. Pursue growth and learning
6. Build open and honest relationships with communication
7. Build a positive team and family spirit
8. Do more with less
9. Be passionate and determined
10. Be humble
CLOUD



210

Tony says, "Our number one focus is our company culture."
For Zappos.com, a great culture translates into great service.
"We want people who are passionate about what Zappos is about
- service. I don't care if they're passionate about shoes." This
passion has translated to hundreds of stories of how Zappos
employees have delivered unexpected service, including finding
out a place to get food, when the hotel had closed room service.

Scrape
For those that are attempting to transform existing Model One
software companies there are many unique challenges. You have
three fundamental organizational approaches. We’ll call these
Scrape, Buy New and Remodel. Scrape refers to an “all-in”
approach to moving from Model One to Model Six. Buy New
means to acquire companies with Model Six business models
and Remodel refers to adopting a Model Four or Model Five
business in parallel with your traditional Model One business.

The Scrape strategy is best illustrated by Concur. Concur is a
unique case study—a company that made the transition from a
traditional license model to software as a service: from Model
One to Model Six as a publicly traded company. Figure 10.2
shows you the transformation in revenue terms. You can see the
HUMAN RESOURCES


211
license revenue line marching to zero, along with the growth of
the subscription line. If you were to look at the stock price for
Concur when CEO Steve Singh made the announcement that
they would shift business models, you’d see the price crashing
below $1 a share, so this is not a change that’s easy to make.
CEO Steve Singh says Concur was able to do this because the
2001 Internet Bubble burst gave them an environment where no
one wanted to own technology stocks. Of course, with the
global meltdown of 2008, maybe we are entering another
window of opportunity.


Figure 10.2 Concur P&L

The Scrape strategy can work, but the transition will not be
easy. During the shift, Concur went from 700 employees to 300
and replaced the majority of the executive team. Not something
for the faint of heart.
CLOUD



212

Buy New
Buying a Model Six company is another strategy a Model
One company can adopt. In this strategy you’ll be best served
by purchasing companies with products in adjacent markets, or
in with products that sell to customers significantly smaller than
your current customer base. Once acquired you’ll need to be
cautious of the “white corpuscles”, which will organizationally
seek to envelop the foreign group and in time destroy it. Your
challenge will be to leverage the knowledge of the people in
development and delivery of the applications as well as in the
new techniques for marketing and sales. Give them the
opportunity to teach and share their experiences. Consider
making their VP of Operations responsible for operations of the
entire company, or make the same step in sales or marketing.

Perhaps the most famous example is Siebel’s acquisition of
UpShot. In the late 90s UpShot was in head-to-head competition
with Salesforce.com, as anyone traveling through the San
Francisco airport would have seen. Those of you considering
the acquisition strategy might consider the story of UpShot as
told by founder Keith Raffel. These days, Raffel has become an
HUMAN RESOURCES


213
author of a different kind, having just released his second novel
Smasher, a popular mystery set in Palo Alto.
31


In the early 1990s, Raffel was working at a start-up where
there was no good way to send leads out to the field sales team.
Not only that, but when the CEO asked him how effective the
marketing programs were, he would point to all the deals that
could be traced to one campaign or another. The salespeople
would then pipe up and claim that they already knew about any
prospect the marketing team uncovered. So Raffel came up with
a way to distribute leads and send them out to the field using
email and an electronic bulletin board (remember them?).

When Keith saw the first Netscape browser in 1994, the light
bulb went on. Raffel took a leave of absence and started
working on a prototype in 1995, and founded UpShot in 1996.
Their aim was to make traditional CRM tools more accessible
and affordable by using the Web. Raffel started by finding what
he calls “a brilliant engineer, who was between gigs” to build a
prototype. He showed the prototype to ten companies, and nine
said they would buy it. So he built a team that included Kris
Olson to lead marketing and Bob Schulman as head of
engineering, and proceeded to rent space above a Jaguar dealer
in Redwood City. As Olson said, “HP might have been founded

31
Raffel, Keith. Smasher. Midnight Ink, 2008.
CLOUD



214
in a garage, but UpShot was founded over one.” Raffel tells a
story from the early days and gives you a glimpse into the role of
a start-up CEO. “A key engineer came to me one day and said
that the paint fumes from below were leaking into his office. I
went to the hardware store, bought some Spackle, and patched
the walls.”

UpShot became the first CRM software delivered as a service
in August 1999, about the time Salesforce.com was founded.
They signed up Fortune 500 companies like Johnson & Johnson,
Xerox, and HP. But ultimately they needed more money to fuel
the sales and marketing engine (remember what we talked about
in the last chapter). So, in 2003, they sold Upshot to Siebel.
From that point on, unfortunately, momentum slowed. Revenue
for the UpShot product, as we know, is recognized over time and
therefore, unfortunately, Siebel salespeople received their
commissions stretched out over time. It took a year to fix this
problem.

In addition, at the time of the acquisition, Siebel was ready to
launch an internal Model Six product, which was not as fully
featured as UpShot. Nevertheless, Siebel went with what they
had developed internally. More momentum was lost. And
finally, a big New York investment bank wanted to purchase a
HUMAN RESOURCES


215
huge number of seats of the Siebel Model Six offering and asked
for new features. The decision to accommodate them, made
against the advice of the UpShot team, took them even further
off course, and still more momentum was lost. While things
improved gradually, everything changed again when Siebel was
purchased by Oracle in 2006. As Raffel says, “That chapter is
being written by others. So it goes in Silicon Valley.”

Remodel
Finally, Model One companies have the opportunity to create
Model Four or Five businesses to run in parallel with their
traditional business. While we’ve discussed the advantages of
doing this, let’s focus on the organizational challenges you’ll
have. First, there will be conflict on the delivery side with any
premium support offerings or professional services offerings
your company is providing. These teams will say a standardized
Model Four or Model Five service offering is too restrictive and
worse yet will often block access to the customer. By the way, a
well constructed Model Four service will be lower cost, which
will threaten them even more. Finally, these new services will
not endear you to any of your channel partners, who not only
resell your software, but also provide a box and a body. Your
challenge will be to find was at worse to neutralize them, at best
to create new opportunities to package your service as part of a
larger offering. This brings us to sales.
CLOUD



216

Your biggest advantage in the Remodel strategy is that you
can leverage the existing sales force. An application sales
person can either double the size of the deal (application license
plus first year of the service), or competitively out position other
traditional software providers. Probably one of the best stories
from the Oracle On Demand experience was competing head to
head with SAP for an enterprise application sale to the largest
mining company in Latin America. Oracle did not look to have
any advantage, as SAP was both stronger in Latin America and
stronger in mining as a vertical market. But Luis Meisler, EVP
of Latin America, and Alberto Chacin, VP for Oracle On
Demand in Latin America, devised a strategy that won the
business. How?

In a meeting with the mining company CIO we said, “In the
end, an SAP or Oracle application purchase price will be nearly
the same, probably somewhere near a million dollars.” The CIO
agreed. “What have you budgeted to manage the application?”
We were guessing it was about $4M a year, based on the rule of
thumb that the cost to manage the software is four times the
purchase price. The CIO answered $6M per year. So we said,
“In that case, over five years the cost of this application, whether
from SAP or Oracle will cost you $31M: $1M to purchase the
HUMAN RESOURCES


217
software and $30M to manage the software. So, you can buy the
software from SAP and manage it yourself or you can buy the
software from Oracle and manage it yourself. In either case it
will be $31M. However, there is a third alternative: let Oracle
manage the software for $15M.”

The Oracle sales person assigned to the mining company not
only won the business from SAP but also received a
significantly higher compensation, given that the deal included
the license as well as the first year of the services. So you can
see, once your application sales force is compensated for selling
On Demand, you will quickly bring them into alignment with
your goals. Organizationally, you’ll have to create an on-
demand sales overlay team and focus on delivering a lot of
information about how this solution is simpler, cheaper, and
better than doing it the old-fashioned way. Once you have that
core group of reference customers, you’ll be ready to ‘cross the
chasm’.

World of Work
So how much will our work lives change in the future? What
will be the nature of work? Here there are lessons to be learned
from the world of online gaming. World of Warcraft, an
elaborate fantasy computer game, is part of a growing number of
MMORPGs (Massively Multi-player Online Role Playing
CLOUD



218
Games). With over seven million subscribers, at any moment
in time, there are hundreds of thousands, if not a million, people
simultaneously playing the game. What are they doing? While
some are merely spectators, the vast majority are working with
other people—people they know, people they don’t, some old,
some young, in Beijing, in London—all to achieve an objective.
Sounds like the modern World of Work, doesn’t it?

So, how is it that World of Warcraft accomplishes something
that we find so difficult in the world of work? One of the keys
to the success of World of Warcraft is the idea that an
individual's skills are known and non-overlapping. Some
characters have the ability to kill, but not heal. And other
characters can heal, but cannot slay.

Our business world needs to extend these fundamental
ideas in software that brings information and people into a
network. Consider the latest Boeing aircraft, the 787
Dreamliner. The company has pushed outsourcing of
component parts to new levels. In the 787, outside suppliers are
responsible for about 70 percent of the aircraft parts, versus a
previous mark of 50 percent. These suppliers span the globe as
shown below.
HUMAN RESOURCES


219

Figure 10.3 The World is Flat

The world might be flat, but our software isn’t. Business
managers can no longer ‘manage by walking around’, like the
great David Packard taught a generation of HP managers to do.
Today we’re all doing conference calls at 4:00 in the afternoon,
because it’s the best time for Asia, Europe, and North America
to talk. If you want to read an entire book on the implications of
games check out the new book
32
by Byron Reeves and J.
Leighton Read.

Summary
Early in 2007, Abhijit Dubey of McKinsey & Co. led a team
that surveyed CIOs regarding the SaaS model. They learned that
the proportion of CIOs considering adopting application cloud
services in the coming year has grown from 38 percent the

32
Byron Reeves and J. Leighton Read: Games at Work: How Games and Virtual Worlds Are
Changing the Way People Work and Businesses Compete.

CLOUD



220
previous year to 61 percent in 2007. Dubey and Dilip Wagle, a
partner at McKinsey’s Seattle office, have published a report
entitled “Delivering Software as a Service.” There are many
interesting comments in the report, but perhaps the best quote is:
“And finally, they’ll have to build the management processes
and organizational structures to manage two distinct but
potentially mutually cannibalizing businesses (traditional
software and software as a service) with different business
models and requiring remarkably different sets of capabilities.
Although this challenge is formidable, software executives who
continue to put it off risk being left behind.”

We could not agree more.









221

BEGINNING OF SOFTWARE



Congratulations on making it to the end of the book.
Hopefully, you’ve found it enlightening and entertaining, and
you’ll repeat some of the campfire stories. If, as some people
believe, this is the third wave of computing, following
mainframes and client-server then we are indeed at the
beginning of another generation of software as revolutionary as
Excel, Word, Powerpoint, SAP Financial applications and
Oracle databases were in the last generation.

CLOUD



222
This software can take advantage of 1000s of computers, near
infinite storage and a network of mobile eyes and ears that
number in the billions.

For those of you in the traditional software business model,
you have many choices. Do you attempt to build a Model Four
business? Do you acquire a Model Six business, or do three of
your colleagues leave and start a new venture using your
specialized expertise? For those of you already in a Model Six
business, perhaps the book has given you a better understanding
of the challenges your fellow team members face, and you’re
starting to think about how information and community will
change your business in the future.

A few years ago, the CEO of a company that builds sales
compensation software in the traditional model asked me if he
should deliver his software as a service. Of course, I told him all
of the good reasons he should consider doing it. Time passed,
and he called me again and said he’d made the decision and now
that he was well on his way, he asked me what he should think
about next. I told him it has to become more about the
information and less about the software. He said, “Interesting,
most of my customers are starting to ask me—what is the best
sales comp plan?”
BEGINNING OF SOFTWARE



223

Now most of us in the software business are used to being
asked for more features and functions, but here the customers are
asking for something even more valuable: knowledge,
information. Perhaps this is why few of us see Amazon, eBay or
Google as software companies, because what we see when we
when we use their applications is the information: the Pez
dispenser, the best seller, the top 10 places to stay in Tahiti.

We already have pointed out that consumer Internet
applications are all content rich. All of you who are running
software as a cloud service are sitting on valuable business
knowledge. Concur today processes over 10 percent of the
expense reports in the U.S. The information about what hotels,
which airlines, and what times of the year travel is happening is
far more valuable than the transactional efficiency afforded by
delivering expense management as a service. Taleo’s systems
have seen millions of resumes. What is the value of knowing
where there is a deficit in skills, by geography? What if an on-
demand CRM software vendor were to analyze all of the
configured workflows and identify patterns of sales processes
with the shortest sales closing periods? Again, what VP of Sales
would not pay to know what sales processes are the most
efficient in a particular industry or geography? Few people may
CLOUD



224
realize this, but to assess the overall health of the economy, the
United States’ government contracts with Visa and MasterCard
to understand aggregated data of how much each of us, as a
consumer, spends. Again, the value of the information far
outweighs the value of the credit card transaction processing.

Of course, such usage of customer data remains a
controversial topic, as many will be nervous about their trade
secrets being published as best practices. Publication and resale
of identifiable private data obviously violates legal regulations
and privacy laws. But, what rules apply to aggregated
information? When you log on to Amazon, you’re contributing
to their overall knowledge of consumer interests with every
click. Why isn’t this the case for business software?

We could go on and on with new companies and ideas, but
hopefully you get the point. Software is far from a mature
industry. We’re much closer to the Beginning of Software, or
what some call the “Golden Age of Software.” We hope this
book has given you some insights into how to think about the
future of your business, whether you’re part of a large software
company or three people with a great idea in Olathe, Kansas.









225









226
INDEX

A
Acquisition strategy to transition from
Model One · 212-213
Amazon · 17, 72, 75-8, 87-8, 110,
129, 153, 168, 184, 208, 223-4,
226
Animoto · 87, 226
Apex · 114, 226
application cloud service companies ·
28, 31, 34, 42-3, 95
applications
frequency of change · 99
SQL-based · 152, 154
search-based · 154
usage spectrum · 100
weight · 98-9
Aspiren · 61
AT&T · 71-2
availability · 81–83
Azure · 111
management · 69-70, 142-5, 181
management at eBay · 139-141
operator errors · 144
B
Back of the Box Network · 59
backup power generators · 63-4
Barrett, Ed · 148
Beasly, Boyd · 47
Benioff, Marc · 42-3, 116
Bezos, Jeff · 75, 77
billing · 198
Black, Peter · 62
Blackbaud · 52-3, 179, 180, 182
Blueroads · 188
Blum, Chris · 30
Bootstrapping Your Business: Start
and Grow a Successful Company
with Almost No Money
(Gianforte) · 46
Brooks, Fred · 104-6
Brooks’ Law · 105
Brown, Gloria · 29
Bruck, Michael · 150
business models
traditional (Model One) · 3-5
open source (Model Two) · 5-7
outsourcing (Model Three) · 7-9
hybrid (Model Four) · 9-14
hybrid+ (Model Five) · 14-15
SaaS (Model Five) · 15-17
Internet (Model Seven) · 17-18

C
CAL (central application logging) ·
141
Capability Maturity Model (CMM) ·
106
Capossela, Chris · 25
carbon
computer footprint · 91-95
credits · 218
emissions · 67-8
impact · 64, 82
power footprint · 67-8
reduction · 124
INDEX



227
CEMLI (Configuration, Extension,
Modification, Localization, and
Integration) · 11, 126-7
CEMLI framework · 126
Chacin, Alberto · 216
change management · 4, 102, 104,
109, 124, 137-142
channels
indirect · 186-9
Chardon, Marc · 180, 182
churn · 190, 194, 196-7
Cisco · 27
Clementi, Erich · 84
cloud computing · xxx
Cloud Computing Test Bed · 90
CloudFront · 78
CMM (Capability Maturity Model) ·
106
Computer 2000 · 188
Concur · 16, 22, 28-30, 124, 128-9,
200, 210-1, 223
profit and loss · 211
Configuration, Extension,
Modification, Localization and
Integration (CEMLI) · 11, 126-7
Constant Contact · 40-2
consumer Internet
applications · 18
companies · 17
usage spectrum· 100
contracts · 178–79
corporate culture · 208-10
cost of software· 18-22
Crossing the Chasm (Moore) · 204
customer service · 4, 36, 45, 47, 150,
161, 182, 198

D
Dalgaard, Lars · 35-6
data center facilities
availability · 69-70
disaster recovery · 58
power carbon footprint · 67-8
power quality · 63-4
Power Utilization Efficiency
(PUE) 64-5
database architecture · 131
DealerTrack · 51-2
deep Web · 151-2, 154
Delivering Software as a Service
(Dubey, Dilip) · 219
disaster recovery · 12, 58-9, 104
distribution channels · 186-9
Domo, Matt · 78
Dot Dead (Raffel) · 212
Dubey, Abhijit · 220
E
eBay · 17, 19-20, 99, 102, 104, 139-
41, 168, 223, 226, 228, 232-3
availability management · 141
change management · 140
traditional software cost example ·
18-19
EC2 (Elastic Compute Cloud) · 75-7,
79, 228
economics of software, SaaS vs.
traditional · 18-20
EDS · 7
Elastic Compute Cloud (EC2) · 75-7,
79, 228
Ellis, Mark · 83
Ellison, Larry · 30, 43
email marketing · iii, 37, 160, 228
Emerging SaaS Channel, The
(TenWolde, Bibby) · 187
End of Software, The (Chou) · ix
EnergySys · 61
ERP II: best practices for successfully
implementing an ERP upgrade
(ACM) ·138
ERP Trends (Peterson, Gelman,
Cooke) ·138
Estimating Total Power Consumption
by Servers in the U.S. and the
World (Koomey) · 67
eTapestry · 53
evaluation criteria, financial · 200

CLOUD



228
F
Facebook · xiii, 17, 87, 113, 117-19,
163-4, 228
Facebook Era, The (Shih) · 165
facilities, data center
availability · 69-70
disaster recovery · 58
power carbon footprint · 67-8
power quality · 63-4
Power Utilization Efficiency
(PUE) 64-5
financial evaluation criteria · 200
Flock IT · 49-50
Force.com · 113-14, 228
Ford, Henry · 125
free software movement · 5
Front of the Box Network · 59
G
Garrehy, Pat · 121
Gates, Bill · 110, 205, 228-9
Gianforte, Greg · 45-7, 226, 228
Gibbs, Kevin · 111
Gluecode Software · 5
Goldberg, Evan · 129-31, 228
Goldberg, Randy · 34
Goodman, Gail · 40-1
Google · xiv, 6, 17, 26, 49-50, 65, 89,
91, 93, 151-3, 163, 223, 228-9
App Engine · 111-3
Bigtable · 112
Gorilla Game, The (Moore, Johnson,
Kippola) · 206
Gough, Andrew · 80
Graves, Joe · 174
Greenstein, Jay · 36
Gregoire, Michael · vi,34, 128
H
Haarmans, Guido · 121
Haig, Don · 133
Haley, Mike · 80
Harris, Ken · 47
Hatz, Bruce · 34
Held, Jerry · 79
Herrmann, Ed · 166
Hewlett-Packard (HP) · 90-91, 213-4,
219
Hiding in Plain Sight: Google Seeks
an Expansion of Power (Markoff,
Hansell) · 93
Hilson, Scott · 161
Hilton, Michael · 124, 128
HP · 90-91, 213-4, 219
Hsieh, Tony · vi, 208
Hummer Winblad · 201
I
IBM · 5, 8-9, 26, 73, 84, 123, 188,
229
Global Services · 7-8, 182
LotusLive · 26
Rational · 86
indirect channels · 186-7
Infosys · 8, 229
Innovator’s Dilemma · 203
Internet Tidal Wave, The (Gates) ·
110
Iyar, Subrah · vi, 5, 26
J
James, Josh · 39-40, 195, 2293
JasperSoft · 6
JBoss · 6
Jeff Bezos' Risky Bet (Hoff) · 75
K
Karsan, Rudy · 32, 229
Kenexa · x, 31-3, 204, 229
King, Joe · 172
Kippola, Tom · 206
Knol · 153
Konary, Amy · 3
INDEX



229
Koomey, Jonathan · 95, 228-9
L
Leadley, Brenda · 36
Lederman, Laura · 200
Leighton, Read· 219
Leingang, Hank · 178
Lentz, Bob · 37
Lervick, John · 153
Lim, Glenn · xxx
Linux · 6, 12, 20, 76
Lipscomb, Dave · 30
license model, traditional · 9, 13, 210
LotusLive · 26
LucidEra · 170
M
Magic Number · 190, 195-6, 229
Magner, Fred · 127
Manber, Udi · 153
marketing
cost of · 17
email · 160-1
expenses · 4, 232-3
search marketing · 162-3
Massively Multi-player Online Role
Playing Games (MMORPG) · 217,
230
McCoy, Courtney · 29
Meisler, Luis · 216
metrics · 64, 142, 180-2, 230, 232
Microsoft · 25-6, 46, 72, 76, 81, 94,
109-10, 152, 186, 188, 208, 230
Miller, Bill · 41
MMORPG (Massively Multi-player
Online Role Playing Games) · 217,
230
Model Five, Hybrid+ · 14-5, 171, 186,
191, 210, 215, 226, 230
Model Four, Hybrid · 9-16, 22, 52,
101, 123, 172, 175, 185-6. 191-2,
210, 215, 222
Model One, Traditional · 3, 4, 6, 9,
14, 16, 22, 30, 52, 54, 62, 122,
171, 173-4, 191, 193-4, 197, 210,
212, 215
Model Seven, Internet · 11, 17, 54, 83
Model Six, Software as a Service · 9,
11-2, 15-7, 22, 30, 53. 99. 129,
162, 171-5. 179, 186, 188, 191,
193-8, 210, 202, 210, 212, 214-5,
222
Model Three, Outsourcing · 7-10, 80,
178-9
Model Two, Open Source · 5-6
Monthly Recurring Revenue (MRR) ·
xiv, 190, 193
Moore, Geoffrey · 204
Morgan, Priscilla · 192
Morin, Dave · 117
Morrell, Bob · 115
MuleSource · 29
MySQL · 123
Mythical Man Month (Brooks) · 104

N
Nazir, Siara · 161
Nelson, Zach · v, 30, 120
Netflix · 64
NetLedger · 21–22
Netsuite · 21–22, 65
NetSuite · x, xiii, 28, 30-1, 113, 120,
122, 129, 131, 2300
Nippon Telephone and Telegraph
(NTT) · 61-2
No Silver Bullet: Essence and
Accidents of Software
Engineering (Brooks) · 150
Nordstrom’s guarantee · 10, 180-1,
230, 232
NTT · 61-2
Nyirjesy Bragale, Christine · 38
O
O’Neil, Mark · vi, 52
Olson, Kris · 213
CLOUD



230
Omniture · x, 25, 39-40, 195, 230
Open Source Initiative (OSI) · 5
open source software · 5-6
OpenTable · x, 51, 53Openwater
Networks · 154
operator errors · 144
Oracle · v, ix, xiv, 3, 10-12, 22, 28-9,
31, 49, 79, 81, 99, 125, 133-4,
215-17, 230
applications · 11, 103, 126-7, 134,
185
data center · 149
maintenance · 3
product support · 154
Oracle On Demand · 10, 12, 125, 134,
149, 172, 192, 216
OSI · See Open Source Initiative
Ouellet, Martin · 33
outsourcing
Brazil, Russia · 9
India, China, Eastern Europe · 9
Ozzie, Ray · 110
P
Packard, David · 219
Palmisano, Sam · 84
performance management · 136, 145,
180, 182, 204
sales · 15
business · 61
Perry, Todd · 164
Pestana, John · 38
Phillips, Everette · 30
power · 33, 49, 55, 57, 63-4, 66-7, 69,
73, 89, 93-5, 117, 227-9
power consumption · 64, 65, 67, 93,
95
power outages, cost of · 63
Powers, Dave · 80
programmers · 5, 19, 49, 109, 144
Public Data Sets ·78
PUE · 65-6, 92
Q
QA · See software:Quality Assurance
Qlusters · 29
Quan, Dennis · 130
R
Raffel, Keith · 212-5, 228, 231
recruiting · 11, 32, 34, 204
Recruitsoft · 33
Red Hat · 6-7, 231
Reedy, Lynn · 139, 142, 207
Reeves, Byron · 219
reliability · 57, 59, 77, 84, 105, 128,
138-9, 141
remodel strategy to transition from
Model One · 215-6
resellers · 186-7
Restructuring Outsourcing
Agreements: An Indication of
Failure, or a Tool to Increase
Value? (TPI) · 178
revenue · x, 4, 7-9, 14, 17, 27, 32, 38,
40, 43, 46, 85, 187-8, 191-3, 201,
214
revenue recognition · iii, 190-2, 231-3
RightNow Technologies · x, 45-7, 83,
231
risk management · 88, 115, 125
Riskonnect · 115
Rocha, Michael · 154
Ross, Aaron · 176-7, 231
Roth, Katy · 162
Rudin, Ken · 170
Rudman, Rick · 37
S
S3 (Simple Storage Service) · 94–99
SaaS · See Software as a Service
SABRE Group, The · 208
sales
channel · 13, 171
compensation · ii, 182-3, 222
INDEX



231
productivity · 157
sales force automation (SFA) · 49,
198
productivity · 42, 160, 176
teams · xiii, 13, 172, 176-7
profit and loss · 201

salesforce.com ·x, xiii-xiv, 14, 16, 21-
2, 32, 42-3, 113, 149, 164, 170,
174-6, 179, 198, 200-1, 212
SAP · 3, 21, 27-8, 31, 49, 99, 103,
152, 216-17, 226, 231
applications, financial · 100, 221
SAS 70 (Statement on Auditing
Standards No. 70) · 82-3, 148-9,
181, 232-3
Schulman, Bob · 213
Schultze, Axel · 188
Scrape, strategy to transition from
Model One · 210-1
search-based applications
optimization · iii, 151, 153-5
search · 49, 51, 80, 91, 145, 153, 169,
182
security · xii, 4, 10, 39, 70, 83-4, 101,
132, 137, 148-50, 155, 173, 180,
226, 232
Care Rehab example · 148
management · ii-iii, 70, 82, 146-7,
181
Mexico treasury secretary example
· 149
Security Threats from Within (Bruck)
· 150
Service Level Agreements (SLA) · iii,
77, 179, 230, 232
Nordstrom’s guarantee · 180-1
Service Science · 9. 232
Seven business models · xi, 2
Singh, Steve · 29, 30, 211
SLA · See Service Level Agreements
Smasher (Raffel) · 213
social graph· 89, 118-19, 164-5
social networks· 163-4, 169
software
business ·xiv, 198
cost of · 18-22,
economics 21,228
traditional · xi, 19,43, 175, 220
sofware companies · xv, 1, 4-7, 9, 16-
17, 22, 32, 43, 71, 97, 173-4, 191,
197, 203, 210, 223 social
networking · 146–48

software as a service (SaaS) · viii, x-
xi, 56, 125, 170, 182, 187, 227-8,
231-2
specialization · 12, 48-9, 100, 103,
133, 232-3
standardization · 11-12, 15, 103, 114,
137, 233
storage cloud services · xii, 55, 57, 74,
76, 85-6, 95, 97, 107
subscription · xiv, 197-8, 202
specialization · 53, 203–6

Spohrer, Jim · 9
Successfactors · 31, 35-6, 204
Szefler, Marian · 125
T
Taleo · x, 31, 33, 86, 95, 128, 180,
204, 233
technology birth year · 92, 95
technology powered companies · 91
traditional software companies · 13,
16, 29, 34, 131, 155, 167, 171,
174, 191, 201, 207, 233
transitioning from Model One · 29-30,
210-11, 226, 228-9
Twitter · 163-4
Tzuo, Tien · 193, 198
U
Udzinski, Chuck · 37
upgrades · 4, 11-12, 21, 34, 99, 123,
134, 137-8, 173, 181, 199, 230,
232-3
UpShot · 16, 212, 214, 233
usage spectrum · 100, 226, 233
CLOUD



232
V
value added resellers (VAR) · 187,
233
vendor-specific objective evidence
(VSOE) · 191-2, 234
Vertica Systems · 97
Viasite · 11
virtual machines · ii, 76, 80-2, 111
Visualforce ·114, 234
Vocus · x, 33, 37-8, 200, 234
Vogel, Werner · vi
W
Wagle, Dilip · 220
Wainewright, Phil · 114
WebEx · x, xiv, 16, 25-7, 32, 157,
174-5, 182, 200-1, 204, 234
profit and loss · 169–70
Winblad, Ann · xiv, 201-2
World of Warcraft · 15, 217-18, 234
X
XenSource · 29
Y
Yahoo · 17, 151, 153, 163, 234
Z
Zaidenweber, Natan · 116
Zappos · 209-10
Zipcar · 192. 234

© 2010 by Active Book Press 2nd Edition All rights reserved.
,6%1 

Printed in the United States of America

Contents
ACKNOWLEDGEMENTS..........................................................................V INTRODUCTION.................................................................................... VIII SEVEN BUSINESS MODELS......................................................................1 MODEL ONE: TRADITIONAL .........................................................................3 MODEL TWO: OPEN SOURCE........................................................................5 MODEL THREE: OUTSOURCING....................................................................7 MODEL FOUR: HYBRID. ...............................................................................9 MODEL FIVE: HYBRID+ .............................................................................14 MODEL SIX: SOFTWARE AS A SERVICE ......................................................15 MODEL SEVEN: INTERNET .........................................................................17 COST OF SOFTWARE ...................................................................................18 SUMMARY ..................................................................................................22 APPLICATION CLOUD SERVICES .......................................................24 COLLABORATION .......................................................................................25 WEBEX STORY .............................................................................................26 FINANCIAL APPLICATIONS .........................................................................28 CONCUR STORY...........................................................................................29 NETSUITE STORY .........................................................................................30 HUMAN RESOURCE APPLICATIONS ............................................................31 KENEXA STORY............................................................................................32 TALEO STORY ..............................................................................................33 SUCCESSFACTORS STORY .............................................................................35 CRM FOR MARKETING ..............................................................................36 VOCUS STORY .............................................................................................37 OMNITURE STORY .......................................................................................38 CONSTANT CONTACT STORY ........................................................................40 CRM FOR SALES ........................................................................................42 SALESFORCE.COM STORY.............................................................................42 CRM FOR SERVICE ....................................................................................43 RIGHTNOW TECHNOLOGIES STORY ..............................................................45 VERTICAL APPLICATIONS ..........................................................................48 DEALERTRACK STORY .................................................................................51 BLACKBAUD STORY .....................................................................................52

i

.............................82 TD BANKNORTH STORY ...............75 COMPUTE & STORAGE CLOUD SERVICES ...............................................................................................................................................................................................97 APPLICATION SPECTRUM .117 NETSUITE STORY .............................................................................................................129 ii .....................................................................................................................54 SUMMARY ....................................91 SUMMARY .......69 SECURITY MANAGEMENT .....87 PARALLEL APPLICATIONS ........................71 ECONOMICS....................................................................................................................57 NETWORK CLOUD SERVICES ..............................................................................................90 TODAY’S CLOUDS ARE SMALL ..................................................................................................................................80 SECURITY MANAGEMENT ....................................................................................................................................................74 COMPUTE & STORAGE CLOUD SERVICES ...................................................................................................................................................61 POWER QUALITY .......................................................................................106 MICROSOFT AZURE STORY .......................................122 CUSTOMIZATION ........................................................................................67 PG&E STORY .....................................................................86 PEAKY APPLICATIONS ........................................................................................................................104 HORIZONTAL PLATFORM SERVICES .....................................................72 SUMMARY ........................................................70 AT&T STORY ................................................................................................................................................................................................................................................................................................................................120 TESTING ......................................83 PUBLIC AND PRIVATE CLOUDS .................................................................................................................COM STORY ..........88 HP STORY ..........................................110 GOOGLE APPENGINE STORY ................................................................................................................125 TRADITIONAL VS...................................................58 NTT STORY .......................................63 POWER UTILIZATION EFFICIENCY ...................OPENTABLE STORY.....................................................95 PLATFORM SERVICES FOR SOFTWARE DEVELOPMENT........................................................................................53 CLOUD SERVICE STACK ..........................................................................98 MULTI-TENANCY.............................................................84 HIGH GROWTH APPLICATIONS .......................................................83 IBM STORY .......................91 COMPUTER CARBON FOOTPRINT .....113 FACEBOOK STORY ................................................64 POWER CARBON FOOTPRINT ...................................................................................................111 VERTICAL PLATFORM SERVICES ..................... CLOUD DEVELOPMENT .............................................................113 SALESFORCE....................56 DATA CENTER....76 VIRTUAL MACHINE .....................................................................101 SOFTWARE DEVELOPMENT LIFECYCLE .........69 AVAILABILITY MANAGEMENT ..........................................................................................................................................................

.........................178 SERVICE LEVEL AGREEMENTS ..............202 HUMAN RESOURCES............................................................................................................161 SEARCH MARKETING ...................................................167 SUMMARY ...........................................................142 PERFORMANCE MANAGEMENT ..........................................................131 PLATFORM SERVICES FOR OPERATIONS ......162 SOCIAL MEDIA MARKETING ......................................151 SUMMARY ............................203 iii .................146 CARE REHAB STORY ................184 INDIRECT CHANNELS ............................................................................133 CHANGE MANAGEMENT ...........................................................182 INSTALLED BASE SALES............196 FINANCIAL SYSTEMS................................................173 CONTRACTS ...........................................137 EBAY STORY ........................................................................................................................................157 TRADITIONAL SALES & MARKETING ..200 SUMMARY ..............................................................................................145 SECURITY MANAGEMENT .................................................................................149 CUSTOMER SERVICE ................................................................................................................179 SALES COMPENSATION .....................................197 FINANCIAL ANALYSTS .........................................................158 EMAIL MARKETING............................................................................................163 GET NOTORIOUS ................190 REVENUE RECOGNITION ..........................................................194 CHURN: COST OF KEEPING A CUSTOMER ....................................................................................................................................................................................................................200 VENTURE CAPITAL...............................................................................165 EDUCATE & SELECT...170 TRANSFORMING TRADITIONAL SOFTWARE SALES .......................................................................................................................................................................................191 MONTHLY RECURRING REVENUE (MRR) ...148 MEXICO STORY ..................................................................................................SUMMARY ............................................................................194 COST OF SALES: WHEN TO HIRE FOR MORE SALES ...........................................................................................................................................................................................160 E-LOAN STORY .....................................................155 MARKETING ..............................................................................................................................................................................................150 SEARCH-BASED APPLICATIONS ....................................................................................................................193 CASH ........................................................................................139 AVAILABILITY MANAGEMENT ..........................................156 ECONOMICS OF SALES & MARKETING ...................................................................168 SALES.............................................................171 SELLING NEW APPLICATION CLOUD SERVICES ........................................................................................................................................186 FINANCE ....................................................................................................................................................

...........................212 REMODEL ............................... RIGHT STAGE ....................................................................................................ECONOMICS.................................................................................................219 BEGINNING OF SOFTWARE ......................207 CULTURE.........................................................................................221 INDEX.........................................................................................................................................208 SCRAPE ...........................................................................................204 RIGHT PEOPLE.............................204 RIGHT PEOPLE ..........................................................................................226 iv ............................215 WORLD OF WORK ............................217 SUMMARY ....................................................................................................210 BUY NEW ............................................................................................................................

At the application level nearly all of the companies who’ve developed business application cloud services have lectured including: Evan Goldberg. It was born of my desire to help both technology students understand more about the world of business as well as to have business students understand more about technology. So in 2005 when I left my role as the President of Oracle On Demand. Zach Nelson. With nearly 50 unique guest lecturers we have spanned the range of companies that are either delivering their applications as a cloud service or providing the platform to do so. so I’d like to thank many of them for their contributions. We just completed the 5th year of the class.Acknowledgements For almost fifteen years I had the pleasure of teaching introductory computer architecture to over a thousand students at Stanford University. I launched the first seminar class on software as a service. with the help of Claire Stager. v . Some of these lectures have been captured explicitly and implicitly in this textbook.

Subrah Iyar. Jeff Jordan. and Erich Clementi. Maynard Webb. whose hard work and perseverance provided me the means to realize many of my dreams. without whom the past twenty years would not have been anywhere near as meaningful or as exciting. Tom Leighton. For this and more I am indebted to my parents. Werner Vogels. And finally thanks to my wife. Greg Gianforte. Danielle. Jonathan Schwartz. Joe Kennedy. Mark O’Neil. Rudy Karsan. Udi Manber. My mother did that calligraphy at the age of ninety. Brian Behlendorf. Steve Singh. vi . Samir Aora. and our three children.Marc Benioff. Mary Ann and David. At the platform and infrastructure level the lecturers included: Charles Phillips. Reid Hoffman. Sue. I’d also like to acknowledge Janet Stevenson. Pradeep Sindhu. We should all be so strong. The class also focused on leaders on the consumer side as well including: Tony Hsieh. and Michael Gregoire. Finally special thanks to Professor Qianchuan Zhao. for her artwork. It was the preparation for those lectures that created the first generation of this textbook. Dave Girourard and Philip Rosedale. and the early readers of the textbook: xxx. Vic Gundotra. Alexandra and Caroline. The Chinese character on the cover is the word for cloud. Marc Chardon. for inviting me to deliver 10 lectures at Tsinghua University in the winter of 2009.

vii .

or for those considering moving existing applications to the cloud. Instead.Introduction Cloud computing. Software+Service. platform as a service (PaaS). In that light we introduce a five-layer cloud services stack to educate you on many of the new cloud services you can buy and not build. we open with seven business models for software and use these models in the discussions throughout the remaining chapters. our mission is to deliver a holistic and balanced view useful for the technology student to understand the business viii . software on demand. infrastructure as a service have all been used to describe new ways to build. or are they similar names for different ideas? This textbook is written as a primer for anyone trying to make sense of what many believe is the 3rd major wave of computing. deliver. The book is written for any business contemplating developing new application cloud services for internal or external usage. this textbook is not written for the experts in marketing to read the marketing section. and client-server computing. and purchase software. Finally. Are they just different names for the same thing. software as a service (SaaS). managed services. Hence. While it contains some aspects of technology the book is first and foremost written from a business perspective. or experts in operations to read the operations chapter. after mainframe.

over a long weekend. In March of that year.challenges and the business student to understand the technology challenges. Suddenly. I thought to myself that. but rather could be written down. abused and often lost. So. Part of the book was devoted to ix . During the meeting he told me he had assigned two salespeople to selling Oracle On Demand. knowledge could stop being just campfire stories that were repeated. and send them back out into the wilderness. hoping they will be able to repeat the stories. the Managing Director of Oracle Germany asked to meet with me as the President of the Oracle On Demand business. perhaps during a long plane ride or over a few evenings. an Oracle salesperson or a potential customer could come to understand why moving to “software as a service” was a logical and inevitable step. of course. I hammered out the story I had been telling around the campfire with the intent that. printed and passed out to many people. but what I couldn’t get out of my head was “what are they saying to customers?” I realized that the way we educate salespeople is to bring them in to the corporate campfire. the good news was that we have two new sales people. tell them lots of great stories. Any student of history knows there was a massive change when Gutenberg invented the printing press. That book was called The End of Software. The origins of the book can be traced to early 2002.

These fifteen include Concur (1999). Most people were debating why ASPs (Application Service Providers) failed and whether SaaS (Software as a Service) would be successful. who all deliver business application cloud services. consider since 1999 fifteen companies. So for those of you who wonder if companies will purchase business application software as a cloud service I think you have your answer. None of these companies was public at that time.8B. Webex was acquired by Cisco in March 2007 for $3. Vocus (2005). Webex (2002). Salesforce. Salesforce. Kintera (2003). RightNow Technologies. Taleo (2005). the world is completely different. As 2009 comes to a close the remaining eleven companies have a combined market cap of over $15B.com.com had generated only $5M in revenue. Successfactors (2007). Constant Contact (2007).com (2004). Websidestory (2004) Kenexa (2005). Netsuite (2007). Blackbaud purchased Kintera in 2008. x . RightNow Technologies (2004). Omniture acquired Websidestory and most recently Omniture itself was acquired by Adobe in October 2009 for $1. have become public companies. DealerTrack (2005). and NetSuite. To get an idea of how different. Eight years later. Omniture (2006).case studies of three small private companies: Salesforce. Some of these companies have been acquisition targets. and Opentable (2009).2B.

each with a different audience. but rather to make sure. Seven Vices or Seven Virtues. Timothy. In this chapter. “Data Center” begins 1 Chou. June/July 2005. “Seven Business Models” was originally published by the SIIA1 (Software and Information Industry Association) and based on a talk Scott Russell invited me to give at Diamondhead Ventures in 2005. xi . The intent is not to argue that one model is better than another. Chapter 3. as a software business. from traditional software to open source. The cloud service stack is meant to further differentiate each layer of cloud services.Chapter 1. to the consumer Internet. We introduce the idea of a five-layer cloud service stack and dedicate Chapters 2 through Chapter 5 to covering each of the major layers. we outline the seven business models that encompass the entire software industry. Anyone contemplating building new applications or transforming their existing applications to cloud services can today take advantage of 1000s of cloud service provided by companies large and small. you know which model you’re trying to succeed in. to SaaS. This layer of the stack is of interest to the business users. Upgrade Magazine. Chapter 2 begins at the application layer with application cloud services. We’ll discuss all of the application areas that have benefited from being delivered as a cloud service – both horizontal and vertical. to outsourcing.

However. The co-location or data center layer focuses on delivering high quality..000 per square foot to build these new data centers require a minimum outlay of $100M. Vertical platform services. while more restrictive allow the developer to leverage the data models xii . In Chapter 5. The next layer of the stack: compute and storage cloud services is discussed in Chapter 4. Pioneered by Jeff Bezos and Amazon the ability to purchase compute by the hour and storage by the month is creating an environment for a whole new range of applications.g. At $1. This layer is of interest to the operations team at any application. Visual Studio. While not a semiconductor fabrication line. platform. Ruby on Rails). “Platform Services for Development” we discuss horizontal and vertical platform cloud services. or compute & storage cloud service business. instead of focusing only on the developer. the investment is not chicken feed. While having compute and storage available on demand is important it’s likely the cloud computing war will be won by winning the developer. Horizontal services are designed to provide many of the development environments we’ve seen in the traditional world (e. All of the current generation of application cloud services have consumed network and colocation cloud services.at the bottom of the stack. these new platform services include much of the operational environments that were traditionally separate. security and reliability locations to house computers and storage.

we’ll talk about how it’s different and why it addresses some of the fundamental challenges we’ve encountered in traditional software development. who put in place the basic machinery that powers eBay. we’ll address some of the major “buts”. As the price points for these cloud services decrease. you’ll need to figure out how to make sales & marketing a more efficient and xiii .supported by applications like salesforce. In Chapter 7 we cover some of the traditional marketing approaches as well as discuss some of the more innovative solutions.com. Developing software in the newer models is not exactly the same as the traditional methods. the challenges are even greater to find new ways to educate the buyers of your services. how do we customize?” The chapter includes an interesting discussion with Lynn Reedy. and operations of the software. such as “…but. Netsuite and Facebook. After building the next great cloud service you’ll be faced with the traditional challenge of how to market and sell the service. Chapters 7 and 8 are written for the non-sales and marketing professional. quickly and inexpensively. Chapter 6. In this chapter. it’s not secure. Whether you’re trying to leverage your existing sales teams or build a new sales team.” or “…but. the management. “Platform Services for Operations” introduces perhaps the most challenging area for traditional software people.

I hope this book will serve to inform you. Chapter 10.” may be at the end of the book. “This is one thing that's really changing in the marketplace. and Customer Lifetime Value) that are important to running a subscription business. Salesforce. but we all should remember that every software business is a people business. whether you’re a business student trying to make sense of these new technologies and business models or a technology student trying to understand what it means to launch a new business. We believe the campfire stories and lessons apply to all businesses. any technology company requires two things to be successful: money and people. stage-appropriate leadership and the right specialist in the right organizational structure are as key as picking the right development methodology or sales compensation structure. We’ll hear from many experts in this area. Having a solid culture. Everybody is a xiv . “Human Resources. All of the public application cloud service companies spend far more on sales and marketing than any other functional area. Churn Rate.com. and Oracle. Google. In the end though.powerful organization. As long time venture capitalist Ann Winblad said. including sales executives from WebEx. Finally. The chapter entitled “Finance” is meant for the non-finance professional and highlights some of the key financial metrics (MRR.

software company. not just the companies in the software industry.” xv .

.

Greed and Sloth. 1 . the Seven Deadly Vices. “the Great”. one of only two Popes to be given the title. the Seven Virtues: Faith. Not only did Pope Gregory define the seven sins to avoid. Charity. defined the Seven Deadly Sins. Anger. Many would agree that managing a software company requires a little of both. or as they are sometimes called. Pope Gregory the Great. Envy.SEVEN BUSINESS MODELS In the 6th century. Hope. Justice. he also included a balancing set of values. Lust. These were: Pride. Gluttony. Fortitude. Prudence and Temperance.

There are successful companies in each of the models. The models are oriented around application software and we’ll start from the traditional world of business software.CLOUD Today. it’s important to understand the seven potential business models before deciding if they are a Vice or a Virtue. the sales guys will never be able to sell it. others have products that operate in multiple models. “We can’t move to a SaaS model.1 Seven Business Models 2 . Figure 1. You will hear some say. “It’ll be great to not have to wonder if this quarter we’ll close that big deal at midnight on December 31.” As you’ll see some companies operate in only one model. as the software industry undergoes a major transformation to a cloud computing model. The seven business models are complete and include every company that makes money from their software.” while others will talk endlessly about the virtues of the model and how.

SEVEN VICES. Subsequently. It’s here that the spending really goes up. sometimes referred to as the software support fee. Why does it cost so much? Amy Konary. of $4. The business that bought the software then buys hardware. Oracle’s maintenance business is over $12B annually. and one of the earliest industry analysts to follow the change to software delivered as a service. or around $800 per user per year. SEVEN VIRTUES Model One: Traditional This is the traditional software business model. This cost multiplier is not limited to complex business software. in our example. Many would point out that even traditional software companies have large subscription businesses. this translates to about $65 per user per month. cites five major areas that cost 3 . at International Data Corporation (IDC). mastered by many companies. In our case study. the customer is charged some percentage.000 per user. The Gartner Group has estimated that the end user can spend up to four times the cost of their software license per year to manage these applications. and hires or redeploys people to manage the software. including Oracle and SAP. say 20 percent. for the update/upgrade rights as well as maintenance. The software is licensed in perpetuity with a one-time price. In fact in late 2009.

the customer will end up spending $1. In our example.300 per user per month just to manage that software. A Model One business is good since. sales and marketing expense. Software support or maintenance is mostly limited to break-fix. in the software company’s interest 4 . problems. can be funded. so called premium support. which often conflict with their partners or internal consulting groups who also provide value added services to help the customer service the software. and all of the profit realized from software support. and change management.CLOUD companies millions of dollars annually. First. access to a Web site of usage information and upgrades. security. It is. with the large up-front purchase of the software. CIOs’ budgets are dominated by spending on managing software. At a well-run Model One business. of course.000 per user.” The implications of this are substantial. performance. which typically can run thirty to forty percent of revenue. Some software companies also provide additional support services. leaving little money on the table for new programs and projects. “CIOs and their departments really focus on five key aspects of the management and maintenance of their computer and software systems: availability. you’ll see license revenue breaking even. while the purchase price of the software may be $4.

a group of programmers came up with open source as a replacement for free software. In the early stages of any software company. Most major free software projects now have corporate backing of one form or another. said “I would much rather have built a software business on the traditional model—it’s a much better business model for the software company. By 1998." an in-flux of for-profit dollars has changed the landscape of free software remarkably. lecturing at Stanford in 2006. In May 2005.” Model Two: Open Source The free software movement began in the late 80s. WebEx founder Subrah Iyar. IBM made an acquisition of Gluecode Software. and many of these companies have become quite adept at working with the community that participates in the projects. and the users who report bugs and suggest new features. that’s why the consumer chooses to buy versus build. creating the Open Source Initiative (OSI) to promote the new term.SEVEN VICES. a privately held company specializing in integration 5 . this is mutually beneficial. Under the name "open source. SEVEN VIRTUES to move customers forward on new releases so as to not incur expense in supporting multiple releases of the software. in many ways. as the customer needs and wants future functionality and.

Support and maintenance. install. John Roberts. announced the acquisition of JBoss for a reported $420M. and potentially use your software is far better than having them download a brochure. In early 2008 Sun Microsystems surprised everyone with their $1B acquisition of MySQL. in June 2006. Over the past many years. which has been the lifeblood of Model One software companies. A year later. the world's best-known Linux distributor. at least $500M has been invested in a variety of open source software companies. All of this has not been lost on the investment community. The purchase price was estimated at $100M. the founder of SugarCRM has shown a Google map with the heartbeat from hundreds of computers in Europe running SugarCRM software—and SugarCRM has no sales force in Europe. MySQL estimates there have been 100 million installations and over 60 thousand downloads per day. While having people download. Why? The open source model (much like Google) has reduced the entry cost to “try it” to zero. JasperSoft estimates there have been nearly two million downloads and over 20 thousand corporate deployments in 200+ countries of their Jasper Reports. Red Hat.CLOUD software based in El Segundo. open source software companies have struggled with the business model. CA. has been 6 .

300 per user per month the customer would be 7 . EDS (purchased by HP) and IBM Global Services have each built large. other than Red Hat with their Red Hat Network. and gold support has also been difficult to sell. companies have emerged who are eager to take on the challenge.SEVEN VICES. Traditional bronze. silver. Some are providing additional functionality in a paid-for subscription service not available in their free version. But with lots of money and smart people. support fees are still paid. In Model Three software is still licensed from the software company on a perpetual basis. Unfortunately. Recognizing that the cost to manage software systems is much larger than the purchase price of the software. SEVEN VIRTUES difficult to monetize. Some open source companies are experimenting with providing both appliances and an on-demand solution as a way of providing a higher degree of service than traditional software support. Model Three: Outsourcing. revenue has been elusive. it may not be for long. multi-billion dollar businesses outsourcing the management of the software and systems. but now the Model Three business will take on the management of that software for less than the $1. since there has been little value-add other than changing the hours you will answer the phone and how quickly you’ll respond to a request.

over time. ADP is probably the granddaddy of this business model. Tiny Infosys with over $3B in revenue has traded at more than eight times revenue. First. bringing them to 8 . China or Eastern Europe. is also not sustainable. IBM Global Services generates over 50 percent of IBM’s revenue. Second. EDS at the time it was acquired was over $20B in revenue. Of course today every Model Three business is leveraging a global pool of talent whether in India. IBM has recognized that they could grow their Global Services business by a factor of two.CLOUD spending. there are two key challenges. in 2007. massive differences in labor rates are not sustainable. a multiple most software companies would be glad to have. was $50B. including entire business processes. a new group of companies. have emerged. These companies continue to grow and take on higher value functions. While Model Three is good. the dependence of system availability on people. such as Infosys. As much of the cost of the software management is buried in the cost of people. Outsourcers can manage the software either at the client’s site or at their own locations. which. using offshore resources in India to lower the overall cost. with the inherent risk of human error. Model Three can grow big companies and high value companies.

they would erode their margins. Fortunately. SEVEN VIRTUES over $100B in revenue. and an installed base. IBM will need to invest in technology.SEVEN VICES.000 per user with a support fee of 20 percent or $65 per user per month. or unfortunately. The big difference in Model Four is to give the purchaser the choice of having the software company service their own software. “One grand research challenge for us is how to scale up services and how to invest to get year-overyear improvements. In our example. For most traditional software companies. India. distribution channel. Spohrer has said. This is the model Oracle pursued 9 . the Hybrid model. represents a bridge from Model One to Model Six: This business model attempts to leverage what an existing software company already has: products. Could a Moore’s Law of service science be possible? Model Four: Hybrid. Jim Spohrer founded an initiative inside IBM called Service Science. and China—so unless there are highly skilled resources in Antarctica waiting to be discovered. While the opportunity is there. In Model Four the traditional licensing model is retained.” Computer science has benefited from Moore’s Law (where the capabilities of computer chips double about every 18 months). Russia. Model Four. that means the software is still licensed for $4. they are already in the low cost countries—Brazil.

having a Model Three business provide the service for less than $1. From the consumer’s perspective. What’s new is the customer is given an option to have the availability.300. security. the purchase of the software and support remain the same. performance. there was a “Nordstrom’s guarantee”. In Model Four.300 per user per month. It’s important to note this decision enabled revenue to be recognized on a monthly basis. Oracle guaranteed if for any reason the customer were unhappy with the service. We’ll come back to answer the question as to why was it so inexpensive. but first let’s make sure you’re clear on the model. Customers ranged from middle market to Fortune 500 companies and spanned the range of applications from financials to HR. the customer had a one year contract that could be cancelled with 30-days notice. or having Oracle service their own software for $150 a month.CLOUD beginning in 1999 and resulted in the fastest growing business inside of Oracle in 2004. manufacturing. With Oracle On Demand. no questions asked. 10 . there would be a 20 percent rebate in that month. the choice was between servicing the software themselves for around $1. and rather than have a complex Service Level Agreement (SLA). and CRM and supply chain. and changes managed for an additional fee.

email. While there was a great deal of flexibility versus what you will find in either Model Six or Model Seven. process or discrete manufacturing). customers were also given the flexibility to choose where the computers would be located—either at an Oracle owned data center or at a location of their choice. As a results there needed to be the flexibility to customize and integrate the applications. recruiting) customers of Oracle applications implemented much heavier weight processes (e.SEVEN VICES. which could be separately priced and give a customer the choice of how much they valued their customizations.. Additionally. This started 11 . Model Four also allowed customers to choose when changes would be made to the application and when upgrades to major releases would occur. which isolated changes that were made for one customer. There will be more on this later as well. @C (for at customer) means a site the customer owns. SEVEN VIRTUES Unlike some lighter weight business processes (e.g.1 illustrates this flexibility by using the symbols of @H (for at home) meaning a data center owned by the business providing the service. There will be more on this in a later section. there was also a great degree of standardization. sales force tracking.g.. An innovation was provided called CEMLIs. Figure 1.

ultimately computers could be put to work automating the key business processes in the business of servicing software. they always followed the same process.CLOUD from the hardware and moved up to the disaster recovery process. By the way. A simple example is upgrades. Specialization allows for repetition. Standardization allows for specialization. and. disks were network attached versus SAN attached. yes. Oracle On Demand upgrades were done hundreds of times in a year by specialized teams. the key to quality and skill in any area. this principle is followed in spades by Model Six and Seven businesses. 12 . Automation ultimately results in both lower cost and higher quality service. product level margins could be achieved because of a high degree of standardization. while you could choose when upgrades happened. all applications were run on a Linux multi-tier grid. With repetition based on specialization. For example. Which gets us back to—how was it possible to offer the service at significantly lower prices than a traditional outsourcer? Could this business be operated at “product level” margins? The quick answer is. and increasing parts of the upgrade process were automated. whether managing computers or doing heart surgery.

the entire installed base of customers who had already bought the software was open for business. In addition. since the sales teams were compensated not only on the sale of the license but also the first year of the service. There were many situations where there were no traditional software sales persons assigned to customer accounts that had bought a lot of software. any model is not without its downsides. Finally.SEVEN VICES. Alternate channels of software distribution do not welcome Model Four. because they had bought all the software they could eat—an interesting paradox. the ability to offer a complete solution from one vendor was often a significant competitive advantage against traditional software companies who had to go to partners for the solution. from a sales perspective. The second downside is the model can only take you so far from a cost 13 . The model is not cannibalistic. since the traditional license model is preserved. Of course. In fact. SEVEN VIRTUES The advantage of this model is that the software does not have to be re-engineered. you can use your existing sales channel. And. They have made a living selling “boxes and bodies” and see the vendors’ entry into this space as competitive. by having a model that split the software license from the service. the average selling price nearly doubled for on-demand applications.

which is probably closer to $500 per user per month. then you would no longer stay in Model Four. This of course is the story of Siebel and Salesforce. the revenue that is recognized in the first year is less than in Model One. of providing this service is in the neighborhood of $50 per user per month. While Model Four allows for the delivery model to change.CLOUD perspective. which compares favorably with doing it yourself or having an outsourcer provide the service. The cost. this would equate to $110 per user per month. but midstream in the second year this begins to change.com. Model Five: Hybrid+ Many people see the SaaS model as a change in the way software is purchased—in particular. But if someone could provide 80 percent of the function of your product at $5 per user per month. not price.000 per user license fee. If you then add the $65 per user per month for support and the $150 to manage the software results in an “all in” price of $325 per user per month. By the third year. Of course. More on that later. the elimination of the upfront license fee. it is easy to see you are not far from moving the business model as well. if one were to assume a 36-month time to pay off the $4. this model is resulting in much larger revenue stream than was true in the traditional 14 . In our example.

has adopted this business model. The choice of moving from Model Four to Model Five is strictly a business choice. Callidus Software. SEVEN VIRTUES model. if not all. not a technology choice. Rather than split the license from the service as we have seen in Model Four. they are providing one price for the entire service. repetition. As a business model. Since Leslie Stretch became CEO in December 2006 the company’s subscription revenue as a percentage of total revenue has nearly doubled. even MMORPG (Massively Multi-player Online Role Playing Games) like World of Warcraft have adopted Model Six. 15 . the cost to manage the software can be driven down through standardization. a provider of sales performance management solutions. Model Six: Software as a Service As we discussed briefly in Model Four.SEVEN VICES. of the new business software companies started in the late 90s have adopted this model. representing nearly 40% of total revenue. and automation. but there is a floor. Most. If you get to start from scratch and engineer an application for delivery as a service (and only a service) the cost structures drop by another order of magnitude.

This has typically been done in the private and public markets. and being able to plow that cash back into sales and marketing. 16 . as we’ll see. the cost and. Unfortunately. such as the Microsoft channel. While these new companies’ offerings had far less functionality. Model Six is not without its challenges. simplicity of purpose allowed them to compete very effectively. rather than being able to close million-dollar deals for software licenses in Model One. one can debate whether it’s a better model for the software company. UpShot and others.CLOUD Software engineered for Model Six will have dramatically lower cost structures than software as a service offered by traditional software companies. companies in Model Six have to find a way to finance the early acquisition of customers. are at best ambivalent to the Model Six players. as in Model Four and Five. WebEx. While it is clearly a better model for the consumer of the software.com. Siebel had to face the direct challenge of Salesforce. and Concur you will find it took somewhere between $50M and $100M to establish the distribution channels required to achieve their meteoric growth.com. If you analyze Salesforce. Furthermore. With no easy distribution channel. the existing channels.

or Zappos by embedding it in the sales transaction. Amazon. it 17 . Google. eBay. Model Seven: Internet Model Seven encompasses all of the companies in the consumer Internet.com. You only need to look at eBay’s or Google’s financials to realize that there is a healthy spend on sales and marketing.SEVEN VICES. Model Seven is distinctive in a few ways. Google by the ad. You don’t buy books. it’s the advertisers that are paying for it. auction cars or find your perfect mate on Google. the software. Google. Of course that does not mean Model Seven companies have no cost of sales and marketing. Why? Again. First. We all use Google for free. Facebook. each of the companies in Model Seven has adopted business models that are asymmetric. and Amazon. Whether it is eBay charging by transaction. The second big difference is that each of these solutions is highly specialized. SEVEN VIRTUES many companies in Model Six reach $10M in revenue and have no ability to grow larger. Yahoo and eBay are all software companies. Asymmetric business models mean that the user of the software is not directly paying for the usage of the software.com have all been very focused on a single function. these companies have chosen unique and often indirect ways to monetize their intellectual property.

but. and I wanted to expand to Melbourne. if I built some interesting software in Sydney. so no new company could get in the door. Australia. but the cost. much more. the cost of software is much. Cost of Software If you went to work for Dell you'd be well educated on the cost-or economics-of hardware. 18 . Some will say the cost of software is the cost of the CD or the manual. That model resulted in the classical software company “going horizontal”—meaning they began to offer more and more parts of the solution. and difficult to integrate—no one needs to spend much time comparing a traditional enterprise application to consumer Internet applications to see the difference. In the traditional world. Of course. This has led to massive suites of applications. this all shows up in the speed of adoption. Unfortunately. so what is the cost of software? I'm not talking about the price of the software. of people to adopt—clearly not possible if the application tried to do everything.CLOUD comes to cost. The simplicity of the Google interface begged for millions. I had to hire a sales/support engineer and a salesperson. all the time promising full integration. My cost of distribution meant that I needed to extract the greatest degree of value out of each of my customers. if not billions. in fact. difficult to install. these applications tend to be complex.

wait a minute.” so a million CDs would go out in the Sunday paper. But the first question the CEO would have asked would be. I got a new feature I want everybody to see. got a really great idea. “We're software guys. What would have happened on Monday morning? My guess is the 1-800 number would be ringing with calls: “Hey. They'd have just gotten out their C++ compilers and ultimately created an auction product. boss. and we can solve this. boss.' Let's put it in the next release.” we would say.” Then we'd hire a support team. I got the plan. It's called 'Buy It Now.” “No problem. We would have hired a bunch of smart programmers. Monday morning would 19 .SEVEN VICES. SEVEN VIRTUES I'm going to use eBay to help explain how the cost of traditional software differs from that of delivering software from the cloud. “Hey. and my ShockWave DLL doesn't run anymore. right? We would insert the CD in the next Sunday paper. “Okay. Let's put it in the Sunday paper. A few months or years later. how do we get it to anybody?” Some smart VP of Sales and Marketing would have said. This really ruined my Windows 95 configuration. “No problem.” No problem. and so on. Let's assume ten years ago we wanted to start a company building auctioning software. set up a help desk. some bright product manager would show up and say.

The cost for delivering the software. There is no way eBay could possibly have existed under this software cost model. that microprocessor stuff is cool. In the beginning of the 80s. we'll move them to India. We'll hire a bunch more support guys. The problem is economic. Burroughs. We've been doing this since the beginning of software. “Well. but it's 20 . and Data General dominated the scene and companies like Intel were dismissed as not delivering real computers.CLOUD arrive and the first phone call would start with “My Linux server just fell over after I installed your software.” No problem. While you could have built a small auctioning software business. If you think about what has transformed the entire hardware industry. all of us in the traditional software business know how to do this. It has completely transformed the hardware business. and managing it would have been so prohibitive eBay would have had to charge hundreds of dollars for every auction. it has been a relentless pursuit of taking cost down by an order of magnitude every decade. Problem solved. distributing it. Now. What would have been the cost of that scenario? It would have been enormous. hardware companies like Digital Equipment Corporation. no way you would have been a multi-billion dollar business.

If you go down 101 to Mountain View and estimate what it costs Google to deliver their service. So what do the real numbers look like? If you take enterprise software-say from SAP or Oracle-and you deliver it as an ondemand service. you're going to do really well if the cost (not price) is $50 per user per month. forget about it. if the customer is running this software in house. so even if SAP or Oracle charges $150. But that's not the end of our story. guess what that number is?-about $7 per user per month. it's significantly lower than having each and every customer of the software solve the problems uniquely. and if you want to control when you upgrade.” We all know the end of that story. chances are. you're going to have to standardize the hell out of everything. so you'd do really well to get it out there at $50. if you become a student of Salesforce. and analyze their cost (the cost of delivering the software). you'll come away with a number closer to $0. it's costing them closer to $1000 per user per month. this is for a fixed set of functions. That's impressive.SEVEN VICES. SEVEN VIRTUES never going to amount to much. Remember. of course. Now.com. Bottom line: anyone in the software business needs to become a student of the economics of software.70 per user per month. By the way. Now remember. 21 .

com did. and have some of your product line in Model One. You can choose to acquire Model Six companies. this shift is much more than technology. You can choose to move completely to Model Six. we will be at the end of software. You will have many choices. which is what Oracle did. one thing is clear—the traditional software model is ending.CLOUD Summary While it may not be clear to you which model to choose. which is what Salesforce. You can choose to start out a new business in Model Six. which is what Concur did. With increased competition and the dynamics of the open source movement. And. the days of million-dollar software licenses are clearly over. since unlike the change to client-server. Whatever you choose to do you’ll need to think about how the choices affect the entire company. with increased pressure on top line growth without fundamentally altering the cost structure of software companies. which is what Blackbaud did. In the next chapter we’ll drill down a little more into some of the 22 . some in Model Four and some in Model Six. You can choose to use Model One and Model Four.

SEVEN VICES. SEVEN VIRTUES application areas and companies that have been successful in delivering their applications as a cloud service. 23 .

We’re going to focus only on application areas. you’ve been talking a lot 24 . I was having lunch in early 2007 with an old friend Tom Kucharvy. In this chapter we’re going to introduce you to a variety of both horizontal (everyone can use) and vertical (specialized to a particular industry) applications.APPLICATION CLOUD SERVICES It’s far easier to talk about business models and the new cloud computing services from the point of view of an application any of us would use in a business setting. He said. Tim. “You know. which have companies who have succeeded in bringing their companies into the public market.

Collaborative software or groupware includes features such as email. calendaring. your friend at another company is not significantly different. Microsoft Office chief Chris Capossela predicts2 that 50% of 2 Fast Company. but the remaining eleven have a combined market capitalization of over $15B. Collaboration Of all of the application areas. instant messaging. Microsoft. I went back to 1999 and looked up nine (at the time) of the business application software companies. Omniture). so I did a little homework. October 28. of course delivers email (Hotmail) but more recently introduced Exchange Online and SharePoint Online. and who all have gone public since 1999. “It’s happened!” So let’s start our discussion of what areas have seen success. Kintera. 2008. Since then some of them have been acquired (Webex. conferencing.APPLICATION CLOUD about this software as a service. Websidestory. As a result some of the more successful application cloud service companies have started out in collaboration. 25 . who all deliver their software as a service. perhaps the simplest (from a business point of view) is collaboration. wikis and simple project management. so it’s safe to say. The granddaddy of all collaborative software. Now when is this going to really happen?” His question got me to thinking. software on demand thing. The way you use email vs.

Microsoft introduced NetMeeting and ended Quarterdeck’s interest in the 26 . calendaring. They might be able to achieve this.CLOUD Exchange mailboxes will be online within five years. presentations and text documents. FutureLabs was acquired by Quarterdeck. Google has brought its considerable resources to bear by introducing Google Apps. This. Webex Story Min Zhu and Subrah Iyar founded WebEx in late 1996 as a re-start of another company called FutureLabs. But perhaps the company that did the most to prove that collaborative application cloud services could be a high growth standalone business is Webex. In 1995. IBM. including Coca-Cola and Nokia. It was then that Subrah. of course. and the ability to collaborate around spreadsheets. met FutureLabs founder Min. Within a year of the acquisition. is a major battleground area. FutureLabs focused on providing software-client and client-server solutions for data conferencing. who was responsible for Quarterdeck’s acquisition of FutureLabs. not to be left behind has introduced Lotus Live and continues to innovate in this area. using their collaboration cloud offerings. as there are reports of large companies. which includes mail.

perhaps the first business software any company purchases is a financial application. Oracle Financials or SAP every company large or small needs a financial application. by 2000. Whether an Excel spreadsheet. they were generating nearly $500M. Intuit QuickBooks. and the rest. it took Subrah and Min until 1999 to actually launch the service. continues to deliver one of the largest application cloud services in use today. Fortunately. so they launched WebEx. 27 .APPLICATION CLOUD space. ironically. WebEx generated about $2. was also purchased by Cisco.000 meetings per day around the globe. revenues were $25 million. By their 10th anniversary. Microsoft Dynamics. In 1999. WebEx went public in January 2002. While collaboration is the modern day telephone. With over two million registered users and over 85. is history. which. as they say. they spent a lot of time working with customers such as Charles Schwab and StrataCom. Although they started WebEx in 1996. Iyar and Min saw the potential of continuing as a service provider. During 1997 and 1998.5 million in revenue with about 400 customers.

almost every company has a financial application and second. Netsuite.CLOUD Financial Applications Financial applications record and process accounting transactions within functional modules such as accounts payable. procure-to-pay. financial control and reporting. because that application generally has few users. First. Concur made an initial public offering in 1999 and has grown a large business specializing in travel and expense management. financial analytics. perhaps for two reasons. This area has not been easy to penetrate for application cloud service companies. That being said two companies have been quite successful in this area. governance. 28 . who made a public offering in 2007. Financial applications can help provide cash & treasury management. has mirrored the “suiteapproach” pioneered by Oracle and SAP and instead focused on the middle market. accounts receivable and payroll. and travel & expense management. The most complex and expensive business accounting software is frequently part of an extensive suite of software often known as enterprise resource planning (ERP) software.

a global measurement equipment company. Concur began life as a traditional software company. Senior Business Systems Analyst at Cummins. chairman and CEO Steve Singh made a bold decision to reinvent the whole company and make the transition to delivering their expense management software as a service. with seven of the world's ten largest companies using their services to help them manage the corporate travel and expense process. but in June 2000. Global Operating Manager at Agilent. The stock sank to below $1 a share.000 clients. Since those lean years early in this decade. Concur has steadily held to its mission.. Inc. This was a major challenge. nearly the entire senior management team was replaced. Of course if you were smart enough to purchase the stock at the low point your returns would be over 40x today. few examples compare with Concur. A/P & Payroll Supervisor 29 . particularly since they were a public company. Courtney McCoy. By the end of 2006. and the number of employees went from 700 to 300. By the end of the transformation. a leader in diesel engine technology and Gloria Brown. Concur had over 4. Their customers include Vivek Ahuja.APPLICATION CLOUD Concur Story While Oracle was one of the leaders in moving from a traditional model to software delivered as a service.

and Chris Blum founded the company under the name NetLedger in October 1998. CFO. the company broadened the application to manage both back-office and front-office sales. Dave Lipscomb. with financing from Larry Ellison. The company launched its first online financial application targeting small businesses in 1999. few can debate the success of the bold decisions that Steve Singh and Concur made. With this foundation. Evan Goldberg is Chairman of the Board and Zach Nelson is the company's CEO. a $5B+ medical technology company. While some question whether such a transition from a Model One provider to a leading Model Six service provider would be possible in the public markets today. Netsuite’s customers include Todd Spartz. Nomis Solutions. Evan Goldberg. support and marketing. President and CEO of China 30 . Netsuite Story NetSuite is a leading provider of integrated online software that allows SMBs to manage their entire business in a single application. Bill Ford. The company changed its name in September 2003 from NetLedger to NetSuite and made a successful IPO in December of 2007.CLOUD at Stryker Medical. Everette Phillips. Today. a price optimization solution for the financial services industry.

human resource (HR) applications were an early example of widely adopted business applications. and align a businesses workforce. 31 .APPLICATION CLOUD Manufacturing Network and David Stover. It was founded in 1987 and ultimately acquired by Oracle for $10B in 2004. It turns out this has also been a good area for application cloud service companies including three companies that have made initial public offerings in the past five years – Kenexa in June 2005.C. Beginning with tracking of resumes of prospective employees to keeping record of employees’ compensation & benefits. located in Charleston. a Tokyobased manufacturer of polyelastane filament fibers used mainly in textiles. S. develop.. HR software tracks the lifecycle of many companies most valuable asset: their people. Taleo in October of 2005 and Successfactors in 2007. is a subsidiary of Tokyo-based Asahi Kasei Fibers Corp. Asahi Kasei. Peoplesoft. is perhaps the best known company specializing in HR software. acquire. Human Resource Applications Along with financial applications. You might find it hard to believe but they replaced their SAP R/3 with Netsuite in 2008.. CFO of Asahi Kasei Spandex America. performance appraisals. Human resource applications are built to assess.

But Kenexa’s path has not been without its challenges. Karsan talked about a day in 1991 when he had mortgaged his house. Karsan took the check. maxed out his credit cards. Over the past 20+ years. Kenexa is a provider of software: proprietary content. drove from Philadelphia to New York City. Kenexa took the long way home. and was three months late in the office rental payments. on that Friday. Through these acquisitions. Kenexa has grown its business to over $160M in revenue in 2009. but as a recruiting firm. Kenexa has bought and sold over twenty businesses. the customer would have waited to the last day to pay. but for some reason the checked arrived early. The good news was that the company had sold a $40. the landlord had told him the sheriff would be coming on Monday to evict them. Kenexa is the oldest of the group. His message 32 . On a certain Friday.000 contract to a large insurance company. having been founded by Rudy Karsan in 1987. Today.com and WebEx started out knowing what they wanted to be. cashed the check at the issuing bank. Typically. and was ready on Monday morning with his rent payment. At a lecture at Stanford University.CLOUD Kenexa Story While Salesforce. Kenexa began operations in 1987 not as a software company. services and process outsourcing that enable organizations to more effectively recruit and retain employees.

one million users. you might be as hard working and as smart as the next group. 33 . Taleo Story Taleo traces its beginnings to 1996. Taleo had over 12 hundred customers. This was his lucky story. Taleo’s software is used by Fortune 500 companies as well as smaller companies to match all sources of talent—be they professional and hourly candidates. agency referrals. contingent workers. Tetu understood the value that. or existing employees—to all positions. campus recruits. Following on that success. Taleo joined fellow application cloud service companies DealerTrack. Kenexa and Vocus in going public in 2005. but. success is based on a lot of luck. by leveraging the power and connectivity of the Internet. By late 2007. and had tracked over 66 million candidates. which would become the most successful job board in Canada. in the end. Ouellet founded Recruitsoft in 1998 and joined forces with Louis Tetu. one could deliver a talent supply chain management solution that could provide the same bottomline and process improvement benefits to large enterprises that supply chain management had brought to manufacturing. when founder Martin Ouellet created Viasite.APPLICATION CLOUD to the students was. who had a background in supply chain management.

He sees the challenges of the traditional software business as being interlocked with the technology. the cost for the upgrade becomes prohibitive. Meaning that. today.000 people worldwide. With a schema change comes an upgrade. if a traditional software company wants to deliver any significant function. Global Staffing Director at Applied Materials. the business can’t move forward since the technology can’t move forward. Mike’s lecture at Stanford in 2007 was entitled: “Does Business Evolution Drive Technical Innovation or Does Technical Innovation Drive Business Evolution?” He ended the talk by saying. Christoph Thoma. there must be a schema change.CLOUD whether it is centralized. Some of Taleo’s customers include Randy Goldberg. decentralized. “Yes!” 34 . and Bruce Hatz. and. Head of HR at Roche. or multinational. an $8B+ provider of semiconductor fabrication equipment. which manages over 100. CEO Michael Gregoire is uniquely qualified to see the similarities and differences between traditional and application cloud service companies. as he was an Executive Vice President at PeopleSoft before joining Taleo. a global healthcare company employing over 75. VP of Recruiting at Hyatt Hotels & Resort. given the often-high degree of customization most customers apply.000 rooms in over 40 countries. Hence.

opened a closet and pulled out a huge three ring binder. that’s not the only one. and then pulled out all a set of org charts and said. and furniture that had been specially designed to reinforce that each individual was part of a team. founder and CEO. You are going to be running Portugal in two years!” Lars said: “What? What are you talking about?” The VP came back and said. I also have you in Germany. On graduation Lars went to work for Novartis. skimmed through it. a large pharmaceutical company based in Switzerland. I also have you …” and he started going through more and more org charts. When he went to the VP of HR to inform him of his decision the VP got a key out of his pocket. “Well. no separate offices. “See. here you are. In short order he was headhunted by Unilever. All Lars could think was “Is this is how they manage careers?” 35 . A few years back I went to the SuccessFactors headquarters to meet with Lars Dalgaard. Successfactors is no different. but that is a lower job.APPLICATION CLOUD Successfactors Story Sometimes you can tell a company is different when you walk in the door. The office had no cubicles. Many companies are born on the personal experiences and frustrations of their founder.

In 2007 they made a successful initial public offering. banking and asset management with more than 180. Their solution is deployed to over four million end users in over 30 languages.000 employees worldwide. and is one of the leading global services providers in insurance. 36 . owner of the Sport and Spine Rehab centers based in Virginia and as large as Brenda Leadley. CRM for Marketing Customer Relationship Management (CRM) is a companywide business strategy designed to reduce costs and increase profitability by managing a customer lifecycle from initially marketing to find and educate the customer. a VP of global human resources for Allianz Group. and marketing to make quick. and in 2008 placed number 133 on Deloitte’s 2008 Technology Fast 500. ultimately close business with the customer. from outside the organization) to give one view of the customer.CLOUD Out of that epiphany he founded Successfactors in 2001. service the customer and ultimately deliver more products and service to that customer. The goal is to allow customer-facing employees in sales. customer support. Allianz Group was founded in 1890. informed decisions. Their customers are as small as Jay Greenstein. CRM’s goal is to bring together information from all data sources within an organization (and where appropriate.

Omniture and Constant Contact. media outlets. based in Lanham. In the next three sections we’ll divide up CRM into CRM for marketing. Today they have over 2000 active customers representing organizations of all sizes across a wide variety of industries. and helps communicate with the media and the public. email marketing.APPLICATION CLOUD This area has been very fruitful for application cloud service providers. This group includes products from Vocus. Maryland. podcast interviews. and publicity opportunities. CRM for marketing includes solutions to track PR effectiveness.000 journalists. 37 . Their database contains information about the media. sales and finally service. Vocus provides a proprietary information database of over 800. went public in December 2005 and provides on-demand. Web-based software for public relations. web analytics and soon new innovations in social media marketing. including in-depth journalist profiles. If the software can increase the number of qualified leads for sales then it’s CRM for marketing. sends out press releases. analysts. contact schedules. Their platform monitors news. As a part of their solution. Vocus Story Vocus was co-founded by Rick Rudman and Bob Lentz in 1992. The company.

joined Goodwill Industries International. address the communications life cycle. and gathering Web visitor data primarily for small businesses. Christine Nyirjesy Bragale. all media relationships were tracked and managed in a notebook (no. For a business that is over $2B in revenue that’s no small step. 38 . from identifying key contacts. and other relevant information compiled by its dedicated media research team. Omniture Story Josh James and John Pestana met as undergraduates at a computer science class at Brigham Young University. developer’s tools. not a computer—a real notebook). one of the largest nonprofit organization in North America.CLOUD pitching preferences. By replacing the notebook with Vocus they are able to not only access records of interactions with journalist but also be able to measure the results of news coverage. as international spokesperson and Director of Media Relations. together. After a successful career in television news at ABC. Vocus’s application is used by a wide variety of organizations. to distributing information. to closing the loop with digitized feedback and management analytics. They focused their first business on Web design. When she joined. The Vocus database is integrated with a suite of on-demand modules that.

resulting in a 70 percent increase in conversion rates.APPLICATION CLOUD While it was a good business the cancellation rate for these small businesses was so high they had to constantly run to stay in place. redundancy.com was losing customers in the checkout process. Based on this data. we had to fight through numerous issues. the company was renamed Omniture in 1998. they refined their checkout process from seven down to three Web pages. Using Omniture’s SiteCatalyst software. Luckily. static Web pages. Omniture is used in a wide variety of applications. like security. before people even knew what to call it. Novell can 39 . “In the early days. some large enterprise customers saw the value of their software and it didn’t take them long to shift their focus to providing Web analytics solutions for those larger customers. Novell uses Omniture SiteCatalyst on their corporate intranet to track the use of applications.” Josh recalls the early days when they were offering software as a service. and availability. Omniture is derived from the words “omniscient” and “future. Reflecting the new business focus. Overstock. and dynamic Web pages. In another application.” Today. We decided the best thing to do was to go after the larger companies and then use them as references.

remove them. but don't feel like you're a slave to every piece of advice you get. only the revenue. acquired WebSideStory and now finds itself as one of the gorillas in Web analytics. Some of the advice I got was on target. which had themselves. Goodman raised more than $20M in her first year and launched the company’s first product: a permission-based email tool. I had never raised venture capital. More recently. or adjust resources so that they don’t have to spend money on pages or applications that are not being accessed. it’s rocketing past you— and there is no way to catch up!" Constant Contact Story Constant Contact." 40 . Some of it wasn't. Since your competitors cannot see the bookings. But. Omniture acquired Visual Sciences. it seems as if the business is growing slowly at first. I networked like crazy. by the time you hear it. The company had just seven employees and a handful of clients when Gail Goodman took the reins in 1999. "just like a freight train. Inc. Gail’s advice to would be entrepreneurs: "Get a lot of advice. which he likens to a freight train.CLOUD better expose pages. was incorporated as Roving Software in 1995. so I talked to lot of people who had done that. Josh loves the subscription business model.

beginning with an unlikely category .. It was just a very dark week or two. “It is an incredible feeling to know that people in another country want to be a part of our extended 41 ." Constant Contact’s customers span the range. Bill Miller. Bethany Church of the Nazarene has an active. Reflecting on her years as CEO Goodman said. the company changed their name to Constant Contact and in October 2007 completed its initial public offering.APPLICATION CLOUD By 2004. their communications pastor made it his mission to update the way they communicated their message to the congregation. “My most rewarding business moment … was reaching 100..000 worldwide clients that depend on its Web-based email marketing service to communicate with their customers. the company has over 150. Not only does their newsletter reach 3. writing the shutdown plan." And "my scariest business moment .000 customers and having a really fun party.000 opt-in subscribers weekly but it also reaches people in multiple countries. Since they started sending out their new monthly newsletter. happened in summer 2001 with less than one payroll in the bank.churches. Today. the church has experienced a tremendous increase in the number of people signing up. dedicated congregation the church inspires through innovative and creative communications.

With that vision. and with a team of three software developers. but in a rented San Francisco apartment. has been building software to increase sales productivity. Salesforce.” said Bill. like Siebel and the leading application cloud service company.com. but was funded instead 42 . next door to his home.com in 1999 to pursue a simple idea: deliver enterprise applications on the Web.CLOUD family. Marc Benioff founded Salesforce. not in a garage.com.com Story The largest and most visible player in the application cloud service market has been Salesforce. “Knowing that the news of our congregation and the work we’re doing has a far greater reach through our newsletter than it ever could have otherwise is encouraging and has a positive effect on everyone involved with Bethany. CRM for Sales seeks to standardize the sales process and make it repetitive. Marc founded Salesforce.com. converted leads by month or who is the tops sales rep this class of software has been adopted by small. It received no venture capital investment. Whether it’s having visibility to a lead by status. The company began. the lead conversion%. medium and large enterprises. salesforce.” CRM for Sales In most companies the largest single expense is sales. As a result much of the initial focus of traditional CRM companies.

The company made an Initial Public Offering in 2004 and. personal to you.” says Marc. Marc constantly changed from a business suit to a Hawaiian shirt and debated himself on the merits of the old and new ways to deliver software. CRM for Service Once you have a customer increasingly keeping a customer is defined by the quality of the service delivered.com’s growth has been meteoric and has symbolized the movement to software on demand.. In 2009. Marc gave his company a mission— to bring about the end of traditional software.com check out Marc’s new book3.com Went from Idea to Billion-Dollar Company-and Revolutionized an Industry. There will be a few stories about salesforce.APPLICATION CLOUD by Marc and Larry Ellison. was generating over $150M in revenue. For the insider’s view of the creation of salesforce.’ as well as raises new questions for existing software companies. Salesforce. Behind the Cloud: The Untold Story of How Salesforce. We all know that high quality service is relevant information. Jossey-Bass (October 19. by 2005. M. 3 Benioff. 2009). Everywhere he went he wore a pin that had the word software crossed out. “Our growth validates the belief in ‘The end of software. they were the highest valued of all of the application cloud service companies. From those early days. At a now-famous keynote speech.com throughout the book. 43 .

CLOUD delivered. Instead.1 shows the dramatic shift from manufacturing and agriculture to services over the last 50 years. contrary to conventional wisdom. Whether thru web. GDP (gross domestic product). this shift has not been an increase in the number of kids flipping hamburgers. chat. email or phone service is all about information.1 Growth of the U. Services Economy Today’s services economy accounts for 80 percent of the U. education. IT services. the U. Figure 2.S.S.S. as you need it. This 44 . At the highest level. 70 percent of the service employment involves highly skilled people. healthcare. Figure 2. and software are all industries that don’t manufacture or grow a single thing. Financial services. economy is a service economy. Furthermore.

Figure 2. doctors. One of the earliest companies to focus on customer service application was RightNow Technologies. scientists. hedge fund managers. Montana. RightNow focused on helping customer service organizations improve service delivery using technology. RightNow Technologies Story Founded by Greg Gianforte in 1997.2 United States Census (1996) Services economy jobs include software designers. Based in Bozeman.APPLICATION CLOUD includes executives. RightNow went public in August 2004. lawyers. Figure 2. Gianforte didn’t start 45 . and customer service professionals and all of these jobs are fundamentally different than working on the assembly line in the early 20th century. and administration and management functions.2 shows the results of the 1996 census.

Often more time than it would take to just go sell something to a customer. G. and Microsoft all originally started without VC funding.CLOUD RightNow with venture capital. In RightNow’s case. they doubled their revenues and employees every 90 days for two years before they took any outside money. 4 Gianforte. Money removes spending discipline. the employees retained more than 75 percent ownership after the company raised $32M. but he reminds the reader that Dell. and. HP. If you have the money you will spend it—whether you have figured out your business model and market or not. 2007 46 . even then. Let your customers fund your business through product orders. You sell your precious equity very dearly before you have a proven business model. He published a book4 in which he builds a strong case against relying on venture capital (VC) funding: • Raising money takes time away from understanding your market and potential customers. "Bootstrapping Your Business: Start and Grow a Successful Company with Almost No Money.” BookSurge Publishing. This is the worst time to raise money from a valuation perspective. • • Gianforte knows he is pushing a contrarian view. and makes a passionate case for starting businesses the old fashioned way.

” says Gianforte. tracks. and collaboration modules. companies can more effectively deliver support and service for their customers. organizations are able to get these advantages over on-premise applications. Consumer Services Manager at Black & Decker. “Delivering applications as a cloud service is more than just another way to implement software. The cloud is altering the way enterprise applications are being consumed. Web self-service. a 50 year old natural nutrition company. Using their case management. Some of RightNow’s customers include Ken Harris. not months or years. a $6B+ manufacturer of power tools. assigns. “Customers want solutions that are up and running in days or weeks. a $3B+ leader in electronic gaming and Chuck Udzinski. It fundamentally changes both the relationship between the software vendor and the customer. and the economics of purchasing and owning software.APPLICATION CLOUD RightNow provides a suite of customer service and support solutions that captures. CIO at Shaklee. and manages customer service interactions from initial contact through resolution. Senior Director of Customer Support at EA. and 47 . and they are looking for ways to reduce overall ownership costs. Utilizing cloud delivery models. Boyd Beasly. email management.

and why does specialization matter? Probably the best example 5 Diamond. others could grow food. but we black people had little cargo of our own?” The answer is that. The lucky ones had the right circumstances to do both.CLOUD today we are just in the early stages of this next evolutionary stage for software. Having the ability to store food meant that the society could develop specialization. 2005) 48 . 1 edition (July 11. Being a food producer meant that the population could grow. Some people could make weapons. and Steel: The Fates of Human Societies”.W. some peoples began to produce food either through herding large mammals or growing crops. Diamond’s book centers around answering a question he was asked by his New Guinea friend Yali: “Why is it that you white people developed so much cargo [goods] and brought it to New Guinea.”5 may make you see the world a little differently. due to conditions in a few isolated places in the ancient world. and still others could erect buildings. Norton & Co. Specialization meant technology could develop.” Vertical Applications Reading Jared Diamond’s bestseller. W. Germs and Steel: The Fates of Human Societies. Germs. “Guns. This is no surprise to us in the modern world of work.. Jared. “Guns. but what does specialization mean in software.

Consider the fact that CitiGroup today employs more programmers than SAP or Oracle. it became immediately obvious what it did and. what they needed to do. But if you walk into any ongoing business. If Google had started out to build software that enabled you to buy a book. that if we provide purchasing.APPLICATION CLOUD of the power of specialization is Google. financial. we’ve built all the software a business needs. you’ll discover that the ERP/CRM footprint is less than ten percent of their overall software. A slightly tongue in cheek example of specialized software is Flock IT. HR. to the teams building the software. auction a Pez dispenser and search. we’re sure history would be quite different. and sales force automation solutions. No one had to teach anyone how to use the system. 49 . So how specialized can software get? We’re conditioned to think that business software is about ERP and CRM.

the cost of supporting and selling this software (pre-Internet) would have been so high that the founder could only support customers in his hometown of Brisbane. But post-Internet. he just couldn’t afford to do this. do we have to think this way? On a global basis. Had he wanted to expand to Sydney.CLOUD Figure 2. doesn’t the future 50 . Search Yahoo! or Google and you’ll find thousands and thousands on Flock ITlike companies. Australia. he would have had to hire a support engineer and a salesperson. In the not so distant past.3 Specializations Flock IT is a company specialized in poultry management software. are there not enough people managing poultry to use this software? And once you see this possibility. Unless he had enough capital and a high enough price on the software.

Morgan Chase & Co. pig management.APPLICATION CLOUD look like lots and lots of specialized software: search. DealerTrack Story DealerTrack was founded in January 2001 with the initial financing of $20M coming from three forward thinking financial institutions: J. and poultry management software? There are some many examples of specialized applications including those provided by Blackbaud for nonprofit organizations. it was this much-needed competition. with each lending institution using a different form—cumbersome for both the consumer and the dealer. Ford and General Motors decided to finance a new company. a dark cloud came onto the horizon. They saw whenever a consumer applied for an auto loan.. In their first year. The company began to add more dealers (60 percent signed up in year three) as well as more 51 . RouteOne. cattle management. AmeriCredit Corp. Chrysler. the old system was both slow and manually driven. Dealertrack for automotive dealerships and Opentable for reservation taking restaurants. and Wells Fargo & Company. DealerTrack brought on line over 20 percent of the 22.000 dealers in the United States. but by late 2002.P. If anything created focus at DealerTrack. to the tune of $100M.. auctioning.

Blackbaud Story Founded by a soccer-playing Englishman. and today the company is connected to almost all of the 22.000 dealers in the U. DealerTrack went public in 2005. Blackbaud is the market leader in software for non-profit organizations.” that allows them to continue to distribute new software products. Below is a chart of the growth of that business. he will say there is no other way to run a software business. DealerTrack has established an electronic channel. a real “information superhighway. Their story contains elements of all of the seven business models. First. Chairman and CEO of DealerTrack.S. 52 . they have a large support and maintenance revenue stream as a by-product of selling in traditional Model One. Not only does he get tremendous economic/cost leverage. talks about delivering software as a service. which they are doing at the rate of three per year. which gives him tremendous pricing leverage. When Mark O’Neil.CLOUD products for the dealers. More interesting. but the application cloud service model also gives the company the ability to release new products quickly and nimbly. In addition a few years ago they began offering their traditional products in Model Four.

On the enterprise side Opentable provides restaurant owners with a reservation management system designed to replace existing paper reservation systems. and price range. thru acquisitions in 2006 of eTapestry and 2008 of Kintera. The solution is an integrated software and 53 . Upon completing the reservation you receive OpenTable rewards points.4 Blackbaud’s Four Product Lines Finally. OpenTable Story As one of the newest public company Opentable is a great example of the blurring of the boundaries between consumer and enterprise software. OpenTable allows you to find and make reservations at restaurants based on time.APPLICATION CLOUD Figure 2. On the consumer side. cuisine. which can be redeemed for discounts at member restaurants. date. Blackbaud has two Model Six companies engineered for multi-tenant low cost delivery.

On the other hand the consumer side of the business is a classic Model Seven business. a consulting arrangement occurs to implement the reservation system. customer preferences.CLOUD hardware solution that computerizes restaurant host-stand operations. integrates with the POS system and allows for reconfiguring the tables for larger and smaller seating arrangements. In many ways the business looks like a combination of a Model One and a Model Seven business. Cloud Service Stack Many of these application software companies were architected in the late 1990s. The system keeps track of VIPs. As a diner. But if you were going to start today there is a much larger array of services that you might consider using. We’ve developed a five-layer stack of services beginning with platform cloud services and including compute & storage 54 . Instead Opentable charges the restaurant a monthly fee to be part of the Opentable network as well as a transaction fee that varies depending on whether the reservation originated from Opentable’s site or the restaurant’s site. you don’t pay for the service. Given the relative unreliability of the network connections to the implementation is a standardized PC. Sales are done using the traditional door-to-door approach and once the buyer says yes.

which give you access to computer and storage resources. Microsoft Azure. Google AppEngine or Netsuite’s NSBOS). Amazon Web Services pioneered compute & storage cloud services. Finally all of these services are connected thru network cloud services. or operations staff to manage these application cloud services (e. Postini for spam filtering or Service. co-location services and the original network cloud services.g.. Application Cloud Services Platform Cloud Services Compute & Storage Cloud Services Co-Location Cloud Services Network Cloud Services Figure 2.now for trouble ticketing). power and secure the physical hardware.APPLICATION CLOUD cloud services. 55 .5 Cloud Service Stack Platform cloud services are either service used by software developers to build new applications (e. Compute & storage services use co-location services to house. This organization will serve to frame the next several chapters..g.

56 . These early pioneers had to build much of the infrastructure and software to deliver and sell these applications. or trying to be the next public company the good news is these new cloud services can speed the delivery and lower the cost of getting new applications into the market. The next several chapters will focus on the different layers of this new cloud stack.CLOUD Summary The early pioneers of SaaS/Application cloud services have demonstrated there is the ability to sell and deliver applications in the new business models. What’s exciting today is the creation of 1000s of new cloud services which new providers of applications can now take advantage of. developing one for in-house use. So whether you’re taking an existing application and delivering it as a cloud service.

we’ll 57 . We'll begin where the name “cloud” began. connected to the network. or if you consume application or platform cloud services this chapter is written to highlight the major areas you’ll need to understand. We’ll move on to discussing one of the key attributes in the reliability and cost of any cloud service: power. Staying with the physical infrastructure. which are the basic network connection services.DATA CENTER The compute. If you choose to build your own compute and storage cloud service. storage and network hardware that power all cloud services ultimately must reside in a facility.

Anyone operating an application cloud service will one day need to operate globally. reliably and efficiently. Network Cloud Services Many have forgotten. but the reason we all call this cloud computing is because we always drew the network (and the Internet) as a cloud. Clearly.000 per square foot it’s some of the most expensive commercial real estate you’ll ever see. 58 . we weren’t sure how it worked but somehow bits were transmitted from one point to another. reliable and fast then operating out of 2 centers on the planet could work. By that we meant. However. this is not the case so it will be important that the application deployments are location independent. and will require at least one disaster recovery site. It is here that the world of appliances and application cloud services intersect. If networks were indeed uniformly cheap. entire books can be written on network cloud service so for our purposes we’re going to focus on one valuable idea. At $1.CLOUD highlight some of the key attributes in building highly available and secure facilities and conclude with the fundamental economics of building data centers.

LOCATION. If you want the people managing the software to be working the first shift. Back of the Box Your next question will be where to locate the computers. the people managing the application will be remote from the physical location of the computer. Once you’ve done that.1 Front of the Box. Europe and Asia time zones.1. If you want to provide for a disaster recovery site. Figure 3. then you’ll locate people in the Americas. as shown in Figure 3. Beyond that. you will need to establish a “Back of the Box” Network to connect the computers to the operations staff. and the cost. LOCATION. Companies like Akamai 59 . versus keeping a team in the US on three shifts. you’ll need to figure out where your users are concentrated. you’ll need at least two sites. performance and reliability of your “Front of the Box” Network. LOCATION At some point in the day or night.

As a result. San Francisco and Paris is critical to global deployment. if you’re application needs to run in Chile. the cost of bandwidth might be such that you’ll need to locate the computers in Chile. But. It becomes more a matter of who pays for what. it will be more expensive. it really doesn’t matter where the building that houses the computers is located. In particular. as opposed to the operational technology. this should be no problem if you have architected a location-independent application. as soon as you realize that standardizing the infrastructure of software and hardware to run in Shanghai. as it will be much more difficult to take advantage of spare capacity afforded by other users consuming the same compute and storage resources. Furthermore. in the end. there is a fine line between application appliances. There are. and a multi-location deployment of any application cloud service.CLOUD have built a large business helping companies provide more reliable. some downsides to not being able to keep all of the hardware in a few locations. Again. faster connections. Furthermore. of course. 60 . it becomes only a financing question as to who owns the actual data center.

NTT Europe Online is a wholly owned subsidiary of Nippon Telephone and Telegraph. France. As part of the contract they are delivering a service for 408 local government authorities to benchmark and improve performance. LOCATION. share leading practices. and Switzerland. Today their customers’ ranges from Web sites such as Nissan Europe to specialized software companies like Aspiren. Spain. Germany. contracted with the UK government to provide the National Performance Management Framework (NPMF). NTT Europe Online is trying to change all of that. improve service levels. and report electronically to the central government. deliver efficiency savings. the largest telecommunications company in the world. which focuses on business performance management software for local and central governments. They provide private cloud services from five data centers across Europe: UK. Aspiren. and EnergySys. You can only imagine the cost and complexity of rolling out software to hundreds local government offices in the traditional Model 61 . LOCATION NTT Story While adoption of business applications delivered out of the cloud has lagged in Europe.LOCATION.

performance and compatibility are managed. and Turkey. EnergySys gives their customers the choice of having the application delivered as a cloud service or as an appliance the customers can host themselves. which transports gas to markets in Azerbaijan.CLOUD One world—or should we say—the help desk phone would probably have melted. now says—“I don’t know why all software is not delivered as a service”. Georgia. They also price the appliance the same as the online system—the only difference is a minimum that reflects the cost of a dedicated solution. Using the EnergySys solution they were able to go live in months. CEO of EnergySys. EnergySys provides carbon-reporting and accounting software for the oil and gas industry. But even if the customer chooses the appliance to be installed “@Customer” EnergSys manages the software and insures changes. One of their customers is the South Caucasus Pipeline Company Limited. Peter Black. And with NTT Europe Online’s help perhaps much more software developed and deployed in Europe will follow Peter’s lead. 62 . who started out the business delivering software in Model One.

LOCATION Power Quality Computers and networks cannot operate without power. It turned out their quarterly test duration was for 30 seconds. Available through EPRI). Madison WI. the systems failed. 2001. but during testing there was enough fuel in the line to run the test for 30 seconds. A few years ago. 63 . After two minutes. there have been several efforts to assess the national cost of power interruptions and power quality.6 EPRI. a large data center had to rely on their diesel generators. 6 The Cost of Power Disturbances to Industrial and Digital Economy Companies. Unfortunately there was a clog in the fuel line. quality data centers will be equipped with UPS and diesel generators. and the National Power Lab data also shows that a typical facility will on average have a voltage fluctuation lasting between one cycle to over thirty minutes 38 times per year. and no more. and no matter how reliable your electric grid. During the past decade. of course periodically run a test to see if the generators are working—and make sure you run the test long enough. You should. there is always the possibility you will lose power.LOCATION. (TR-1006274. LOCATION. The Electric Power Research Institute (EPRI) estimates the cost at over $119B per year. As a result. In fact at least 4 times per year the power will be out for a minute or more. even though they had tested the generators every quarter. June 29.

and Bill Tschudi. EYP Mission Critical Facilities 64 . 2007. Evan Mills. Presented at The Uptime Institute Symposium 2007 9 Best Practices for Data Centers: Lessons Learned from Benchmarking 22 Data Centers Steve Greenberg. Their analysis showed data centers could be over 40 times as energy intensive as conventional office buildings. PUE was established by The Green Grid. In another published work9 a team from the Lawrence Berkeley National Labs benchmarked 22 data centers. The Power Usage Effectiveness7 (PUE) metric is defined as the ratio of the total power consumed by a data center to the power consumed by the IT equipment that populates the facility. the Uptime Institute gathered data from many of the 85 members of the Institute’s Site Uptime Network®.CLOUD Power Utilization Efficiency While having high quality power is important. 7 The Green Grid.” Technical Committee White Paper. as we will see the effective usage of power has both economic and carbon impacts. In their sample size the average PUE was 2.58. 8 Data Center Energy Efficiency and Productivity By Kenneth G. They created a metric defined as the Computer Power Consumption Index. “The Green Grid Data Center Power Efficiency Metrics: PUE and DCiE. Rumsey Engineers Bruce Myatt. To give you a sense of the range of possibilities. Lawrence Berkeley National Laboratory Peter Rumsey. Brill . an association of IT professionals seeking to dramatically raise the energy efficiency of data centers through a series of short-term and long-term proposals.

Figure 3. because large Model 7 companies consume large numbers of servers per employee it makes sense that companies like Google are pushing the state of the art. The 22 data centers ranged from a PUE of 1. 65 . Figure 3.0.LOCATION. LOCATION which as it turns out is the inverse of the PUE.5 to 3.3 summarizes the PUE from all Google-designed data centers with an IT load of at least 5MW and time-in-operation of at least 6 months.2: Computer Power Consumption Index for 22 Data Centers (Inverse of PUE) Again. Figure 3.2 shows the Computer Power Consumption Index for all 22 data centers in their study. LOCATION.

and over time their designs have become more efficient. and the facilities themselves are located in different a climate.3: Google Data Center PUE Why is there variation among the data? Power and cooling architectures differ between facilities. Data Centers E and F are two of the newest and have among the lowest values. In addition. 66 . PUE values are impacted by seasonal weather patterns. which influences the PUE performance. Also. Data Center A is the oldest facility of the group and has one of the highest average PUE values.CLOUD Figure 3. and thus the PUE during cooler quarters tends to be lower than in warmer ones. For example. the data centers shown were built at various times since 2005.

S. this is equivalent to about five 1000-MW power plants for the U. 2007. In total power demand.gov/egridweb/ 67 . eGRID is a comprehensive source of data on the environmental characteristics of almost all electric power generated in the United States.7B and $7.2B for the U. Dr. Koomey completed a study on the global power consumption of servers. an amount comparable to color televisions’ power consumption. the other major area of focus is the reduction of overall power and cooling requirements.S. 10 Koomey. When cooling and auxiliary infrastructure are included. and 14 such plants for the entire world.epa. that number grows to 1.10 Not all power is created equal.LOCATION.2 percent. Jonathan. February 15. respectively. 11 http://cfpub. Some power has a much higher carbon footprint than other. Stanford University. electricity consumption in 2005. Jonathan G. For this answer we turn to the Emissions & Generation Resource Integrated Database11 (eGRID). Estimating Total Power Consumption by Servers in the U. and the World. The total power used by servers represented about 0. The total electricity bill for operating those servers and associated infrastructure in 2005 was about $2. LOCATION Power Carbon Footprint With the increasing cost of power and the need to reduce carbon emissions.S.S.6 percent of total U. and the world. LOCATION.

If you know which specific power plant your data center is connected to then that CPF can be used.CLOUD The annual total output emission rate is the measure of the carbon emissions as it relates to the generation output. otherwise eGRID divides the country into a number of sub-regions.4: Sub-region power grids Interestingly the sub-regions have a nearly 4:1 difference in the amount of carbon generated per Watt (40-170 pounds/Wattyear). Units are in lb/MWh and are converted to lb/Watt-year by multiplying by 0. 68 .00876. Figure 3. Surprisingly there are coal-fired plants that are cleaner than some oil-fired plans. If you go thru the database you will find specific power plants that range from 0 to 876-lbs/watt year.

is offering significant rebates as part of a new energy incentive program for computer servers. This is the first-ever incentive rebate offered by a public utility company for servers. or generators and typically will have only single paths for power and cooling. Tier 2 data centers provide 1300 outage minutes per year and again will typically require maintenance of power to require a shutdown. Tier 1 co-location cloud services may or may not have UPS. The Uptime Institute has identified four levels of availability – the best being Tier 4. the lowest level being Tier 1. LOCATION. one of the largest utilities in the United States.LOCATION. obviously one of the important criteria for a co-location cloud service is the availability. LOCATION PG&E Story Pacific Gas & Electric (PG&E). Tier 1 data centers have an average outage profile of 1700 outage minutes per year. Perhaps we’ll see more of this in the future. Tier 3 co-location 69 . customers replacing existing equipment with new servers from Sun Microsystems can receive a cash savings of between $700 and $1000 per server. Availability Management Whether you decide you need to own your own buildings or just rent. As part of PG&E’s Non Residential Retrofit program.

limiting access to areas on a need to access basis. guards may use mirrors to check underneath vehicles for explosives. a 100-foot vehicle buffer zone. Tier 4 facilities are the top of the heap delivering 26 outage minutes per year.CLOUD services move the outage performance significantly and are designed to deliver 90 outage minutes per year. 70 . In order to be able to do this there will be multiple power and cooling distribution paths. Finally. In general the highly secure facilities will have few windows. Surveillance cameras will be installed around the perimeter of the building. Finally for authenticated individuals. All personnel working in the facility should have background checks. at all entrances and exits. There will be few entry points and for data centers that are especially sensitive or likely targets. Furthermore using two-factor or biometric identification is becoming standard for access to sensitive areas of data centers. foot thick concrete walls and retractable crash barriers. Security Management The security of a co-location cloud service is has both a physical and a people component. This brings us to the people side of the equation. and at every access point throughout the building.

of the Internet bubble. When the bubble burst. AT&T acquired USi in October 2006 for $300M. pain management (e. is a good example of a software company that used AT&T’s co-location services. lower back pain). arthritis. and later a victim. Payroll ballooned to 1. USi raced to build expensive data centers to support the wide range of software it offered. and other wellness programs. As a small company there was no way compliance 71 .. Stock in the once publicly held company.400 employees. Inc.LOCATION. ultimately filing for bankruptcy in January 2002. HealthMedia's software application provides online programs that focus on disease management (e. USi was a creation. Fleets of jet skis were docked nearby for employees eager for a spin on the Severn River. reached $66 a share in early 2000.g. USi was hit hard. Since they are in the health care industry.. Pfizer. and UPS provide some of these programs to their employees.g. including the Health Insurance Portability and Accountability Act (HIPAA) and Title 21 Code of Federal Regulations. HealthMedia. LOCATION. LOCATION AT&T Story In many ways. GlaxoSmithKline. Today AT&T’s services over 150 clients with end users in over 30 countries. weight management. Companies like Jenny Craig. the company must comply with several unique requirements. diabetes).

5 you’ll see an IBM estimate of the cost breakdown for the various components. In 2008 Amazon. Figure 3. with a plan for fifteen more.000 sq foot data center at a cost of $100M. In 2007 Microsoft decided to build a 470. GDS-China is in the process of building five new data centers in China. but leveraging the scale and expertise of AT&T they could focus on their software and leave the data center certification to their cloud provider. At ~$1.5 Co-location Facility Capital 72 . Economics Building data centers is expensive.com broke ground on an 116. And just so you don’t think this is purely a US phenomenon.000 sq foot data center at an estimated cost of $550M.000 per square foot these are expensive capital investments. In Figure 3.CLOUD was possible. Sixty percent of the cost is mechanical and power systems.

that power or energy costs dominate the operational costs of a data center.LOCATION. “In the Data Center. LOCATION This brings us to the on-going operational costs of a data center. $300 Cumulative Cost of Operations ($m ) $250 $200 $150 $100 $50 0 Energy Cost Staffing Bldg. Moreover the cost of power can fluctuate by a factor of at least 6x. 73 . Electronics Cooling Magazine. & Mgmt. Figure 3. C. So in the end it may turn out that future data centers will just be value add to anyone generating power. Power and Cooling Costs More than IT equipment it supports”. Maint. LOCATION. That is.6 is an analysis done by IBM which shows what many who have been students of this area already know. R E Tax 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Year Source: IBM engineering estimates. 2008 Figure 3. from 3 cents per KWh in Idaho to 18 cents in Hawaii.6 Co-location Operational Costs Christian Beladay’s analysis12 shows that as of 2008 energy costs alone have exceeded server costs. 12 Belady.

low-carbon high quality power. Utilizing these co-location cloud services the next layer of the stack is to provide compute and storage cloud services. i. providing computers and storage technology purchased by the hour. These data centers will in many ways be the bridges and roads of the next generation of all industries. or the month.CLOUD Summary Network and co-location cloud services represent significant capital investments to build out..e. We’ll discuss this more on the next chapter. but also by the availability of low-cost. 74 . Increasingly the location of the data centers will be dictated not only by the proximity to networks.

But not until the Simple Storage Service (S3) rolled out. The article focused on Bezos’ vision of transforming the Internet retail giant into a cloud service provider. Amazon began another revolution.COMPUTE & STORAGE CLOUD SERVICES The cover story in Business Week in November 2006 was entitled Jeff Bezos’ Risky Bet13. It started with the Amazon eCommerce API and Alexa Web Search services. Business Week November 2006 75 . 13 Jeff Bezos’ Risky Bet. did it start to become a reality and with the creation of Elastic Compute Cloud (EC2).

Rackspace and Opsource.CLOUD Compute & Storage Cloud Services Amazon’s Elastic Compute Cloud (EC2) and Simple Storage Service (S3) were the first compute and storage cloud services. paying by the hour. or Linux images. In fact Amazon’s S3 pre-dates the availability of EC2. Along with compute services are storage services. which allows for latency optimization and high levels of redundancy. Very typically 76 . Compute cloud services allows scalable deployment of applications by providing a web service through which a user create a virtual machine instance containing a wide-range of software. and terminate server instances as needed.15/GB/month).10. Also these new compute cloud services will allow you to choose the size of memory as well as the number of virtual cores. While pioneered by Amazon. Typically these virtual machines include preconfigured Solaris. starting at around $0. Windows. Some services also provide control over the geographical location of instances. A compute cloud service enables people to create. compute cloud services are now available from Microsoft. Storage cloud services provide "unlimited online storage" again in a subscription model (typically $0. launch. all for an increase in the hourly fee.

Jeff Bezos delivered the keynote speech and shared some pretty astounding numbers for S3. and every customer's experience. With growing usage also comes growing sophistication in the management infrastructure. EC2 provided a service level agreement. The cost of online storage is dramatically lower than tape.95 percent of the time. Chief Technology Officer Werner Vogel described some of Amazon's continued investment in reliability and manageability: “We relentlessly measure every possible resource usage parameter. including a peak day when there were 921 million requests. In keeping with their growing sophistication in 2008. if not three different data centers. hence this architecture precludes the need to provide tape backup and offsite storage. The number of Amazon Web Service users has grown from 240.0 conference.000 in 2007 to 540. Many gigabits per second of monitoring data flow 77 .COMPUTE & STORAGE the storage services keep data in at least two. every application counter. a formal commitment that the service will be available at least 99.000 in Q1 of 2009. Amazon Story At the 2007 Web 2. and in the same time period the number of storage objects has gone from five billion in 2007 to over fifty billion.

SimpleDB general manager. and Public Data Sets. St. Newark. video. Hong Kong. Amsterdam. Louis. Miami. call up an API that returns a unique domain name for the content. Amazon has 14 edge locations around the world including Dallas. Matt Domo says. etc. columnar DBMS on 78 . CloudFront caches high-traffic content on the company's worldwide network of edge locations so it's always near the enduser for low latency. Dublin. music. “Amazon SimpleDB eliminates complexity so that developers and businesses can focus on optimizing applications and not administering their database. Seattle. Frankfurt.” Beyond S3 and EC2 Amazon introduced other services including CloudFront. London. then link to it on a Web site or Web application. and Tokyo. applications. Palo Alto. Amazon is creating a partner community that is further extending their cloud capability.CLOUD continuously through the Amazon networks to make sure that our customers are getting serviced at the levels they can expect and at efficiency level the business desires.) into S3. Vertica Systems offers their highly scalable. Customers store the original versions of content (photos.” In addition to their own efforts. SimpleDB. SimpleDB is a new service providing simple relational database service. Los Angeles.

student at Cal he worked on the relational database that would come to be known as Ingres. “The combination of Vertica’s ultra high performance at extremely low price and EC2’s dynamic configurability are enabling business changing analytics that haven’t previously been conceivable. He was one of the first developers at Tandem Computers and wrote the original file system and ultimately he went on to preside over Oracle’s billion-dollar database business. Bringing analytic applications on-line in hours rather than weeks or months and having access to huge resources for varying durations means that what would have been a multi-million dollar capital investment can be turned into a short term.” And he should know. As a Ph. Jerry says.D. opex decision which can be made by line management.COMPUTE & STORAGE EC2. It enables customers to automatically configure in a matter of minutes analytic databases of virtually unlimited size and performance at a fraction of the cost of conventional systems. “This is a huge step forward 79 . Chairman of Vertica. has been a major figure in the evolution of database technology. Jerry Held. small. But these cloud services are not only being used by new startup companies but also by Fortune 500 corporations including Eli Lilly and Autodesk. With that breadth of experience he joined Vertica as Chairman of the Board and has been instrumental in moving them towards the cloud.

” Gough said. each running its own operating system. and now that Amazon EC2 runs Windows and SQL Server. Virtual Machine Compute cloud services are provided as virtual machine instances. A virtual machine allows the sharing of the underlying physical machine resources between different virtual machines. “Autodesk uses EC2 for back-end data processing tasks. In principle Capgemini is running a Model Three business but using Amazon’s compute and storage cloud to offer its outsourcing clients a lower cost approach to pilot projects and IT projects.CLOUD in maximizing our results relative to IT spend. SaaS business development manager at Capgemini. Further signaling the progress they are making was the recent announcement from Capgemini. Capgemini will offer business users Microsoft Sharepoint.” said Mike Haley. The software layer 80 . a senior architect of search engineering. we have even greater flexibility in the kinds of applications we can build in the AWS cloud. an Eli Lilly associate information consultant who uses the service to process research data. Oracle ERP and application development and testing in the cloud. “Every business has a project that never surfaces because 80% of an IT budget goes on maintenance. According to Andrew Gough.” said Dave Powers.

1 shows the server utilization across a typical data center. thereby reducing power. Figure 4. Amazon uses a variation of the Zen open source implementation to deliver their cloud services. cooling and space requirements.1 Server Utilization in Data Centers 81 . Virtual machines are very useful in situations where the server utilization is low as it allows fewer servers to be shared. VMWare is the leading company in providing this type of software. Figure 4.COMPUTE & STORAGE providing the virtualization is called a virtual machine monitor or hypervisor. although many other companies including Microsoft and Oracle are providing their own.

but multiple groups using it). provide fine-grained access control and finally audit the interactions. 82 . . some will differentiate purely on price. others on quality of service and still others by geographic or industry specializations. One area of particular customer concern is security management. role-based user permissions and the ability to track actions by department. HIPAA (for healthcare) and PCI (for payment) compliance further differentiate one offering from another. As a result some vendors are providing dedicated VLANs. and in some cases carbon impact. standards like SAS 70 Type I and Type II. subadministrators (so there can be one master account. Security Management As more compute & storage service providers enter the market. When providing security management at the compute and storage level one needs the ability to robustly authenticate anyone with logical or physical access. even in non-cloud implementations virtual machines are being used for server consolidation as a means of reducing cost. Furthermore since one of the major principles of good security management is auditability.CLOUD As a result.

private clouds. Anyone delivering services in Models Four through Model Seven can make investments in security on behalf of many customers and. they chose the RightNow service. Public and Private Clouds Being able to control security.COMPUTE & STORAGE TD Banknorth Story As a financial institution. and particularly important for TD Banknorth was the SAS 70 report. Some would say that private clouds are just a marketing term to rename a company’s existing data center. makes a more important point. which are subject to regular SAS 70 Level II audits. While this may be true it’s better again to see the range of possibilities as a continuum rather than a black 83 . TD Banknorth’s evaluation of vendors’ offerings had to be more thorough than most. Senior Vice President of eCommerce at TD Banknorth. for example. perhaps. set forth strict guidelines about what constitutes appropriate due diligence in such a selection process. Regulations imposed by the Office of the Comptroller of the Currency. as a result. said they “especially liked the fact that RightNow uses AT&T’s facilities. specification and performance of the compute and storage infrastructure has given rise to the terminology of public vs.” This. Mark Ellis. can arguably provide even higher levels of service than any one customer could on their own. In the end.

like public swimming pools. . However. Public clouds. just like with swimming pools. to head up cloud computing. are specified by the service provider. however. more recently the idea of reserved instances has been implemented where the user can lock away compute capacity for a year or even three years. it’s clearly possible to implement private clouds. an implementation of a highly standardized AS/400 compute and storage private cloud would clearly reduce the cost and improve the reliability for any of their customers. IBM rarely makes moves like this for strictly marketing purposes and in many ways it signals IBM’s view that cloud computing represents the 84 .CLOUD and white decision. IBM Story In February 2009. e..g. Of course within those specifications the providers continue to provide increasing degrees of flexibility and security. if the specification of the cloud service does not fit your needs then. greater degrees of control on who has access to what. With nearly 1000 customers worldwide. Early implementations of the public cloud didn’t allow you to insure there was capacity when you needed it. IBM’s CEO Sam Palmisano appointed Erich Clementi. As an example consider JDA Software’s market for implementation of its AS/400 E3 product.

In short. as especially for public cloud services the buying behavior shifts from one time investments to small portions of recurring (or even non-recurring) revenue. compared to the traditional IT market. compute & storage cloud services to internal or external customers. and composed by three sub-markets: 1. 2. Within compute & storage cloud services a large portion of business will be driven by clients not buying new physical infrastructure. which is growing very slowly (1% for traditional SW and SVCS) or decreasing (-3% for traditional HW). multiples of 10 cents per hour acquired by many individuals unpredictable over the year.COMPUTE & STORAGE third major wave of business computing – preceded by clientserver and mainframe computing. But why? In 2009 Erich Clementi presented IBM’s analysis of the cloud opportunity. they predict the cloud opportunity is substantial ($126B). software and services purchased by enterprises and direct cloud providers to provide application. e. platform. especially commodity x86 servers and storage. Components: $44B. their models predict the market is growing at 28% (2008-2012 CAGR). Infrastructure services: $30B. Here the shift from traditional to new will appear to be even more dramatic. compute & storage cloud services. In 2012. Includes all hardware. 3. Includes all applications and any business process outsourcing built on top of 85 . Business services: $52B. but using public or private services. Includes the revenue derived by providers of platform.g. In effect this is being in the arms dealer business selling traditional hardware and software.

and build social-networking communities.CLOUD application cloud services. to providing a new horizontal platform service like Pangoo. from moving their own global development labs to using a compute & storage cloud service. As an example IBM uses Taleo to provide certain HR BPO services. small businesses and in the growth markets will benefit dramatically from the new economics and the user experience of cloud. High Growth Applications At one level some might see that the difference between these new compute and storage cloud services and the traditional model of buying servers and putting them in the data center as 86 . along with Telelogic Focal Point. for software testing. In the mid. And even beyond LotusLive. As a result IBM is developing and delivering cloud services across the spectrum. an online software management tool. hold online meetings.and long-term solutions for various industry verticals. they have made available Rational AppScan OnDemand. At IBM's size and reach and with their treasure trove of technology anyone interested in cloud computing will need to know what they're up to. developed in IBM’s China Research Lab or delivering application cloud services like LotusLive is a collaboration tool that lets people share contacts and documents.

COMPUTE & STORAGE only an economic choice. Of course it may be easier to purchase and it the service could be lower cost, but to see compute and storage only in this light would be to miss the true potential. Let us suggest three styles of applications that can truly benefit: very high growth applications; peaky applications and parallel applications. The first category of applications, very high growth applications, and look like you won the lottery. Perhaps only one out of a hundred or one out of a thousand applications might see rapid adoption, but when it does the ability to rapidly scale is critical. Consider the case of Animoto. Animoto is an online service that allows you to create video productions simply. You upload your photos, Powerpoint, choose a song and Animoto creates an effects-heavy movie in minutes. When it launched on Facebook, demand was such that it had to increase the number of compute instances on Amazon’s EC2 from 50 to 3500 within three days. “You could give me unlimited funding,” says Adam Selipsky, VP at Amazon Web Services, “and I wouldn't know how to deploy that many servers in 72 hours”14 Peaky Applications This class of applications is characterized by cyclic usage of resources. Typically this might mean computation that occurs
14

Economist - October 25, 2008- A Special Report on Corporate IT

87

CLOUD every day, but not every night; or every week or every month. Typically in this scenario a traditional model requires that you allocate the largest required resource, even thought it might only be used 1/3 of the period. An example of this type of workload is a Wall Street firm with a nightly business process where they upload the day’s market trading data into a storage service (in this case Amazon’s S2), and then run proprietary risk management algorithms. The analysis typically lasts ~10 hours during week nights and during that period 3000 compute instances are used. That demand is an order magnitude greater than the daytime and weekend workload where the number of instances is closer to 300. Parallel Applications As part of teaching the first class on cloud computing at Tsinghua University, Amazon donated $3000 worth of time on the Amazon cloud. At $0.10 per hour it means you could use one instance continuously for 3.5 years, or 100 computers for 12 days or 10,000 computers for 30 minutes. But what would you do with 10,000 computers for 30 minutes? We’ve never been able to ask that question before and the answers promise to be truly exciting.

88

COMPUTE & STORAGE There are only two areas in business computing that have taken advantage of large numbers of independent low cost computers. In the 80s Tandem Computers created systems that allowed 256 computers to outperform the most powerful mainframes of the era in a transaction processing system, which was optimized to enable parallel execution. And beginning in 2000, Google has shown the power of using 1,000s of computers to enable low cost, high performance search. So the challenge is to identify (and parallelize) algorithms that are not as simple as consumer search. One of the chief architects of the Tandem system, Jim Gary, in more recent years turned his attention to what he called the “fourth paradigm” of science15. The first three paradigms were experimental, theoretical and computational science. He explained this paradigm as an evolving era in which driven by our ability to capture a huge amount of data – whether that’s about the human genome or the interlocking web of 300 million people’s social graphs. Jim’s vision was that eScience could unify theory, experiment, and simulation with a new generation of applications specifically designed to manage, visualize and analyze this even deeper Web of data. Based on low cost compute and storage services Jim saw a world in which all of the

15 “The Fourth Paradigm: Data-Intensive Scientific Discovery.” Edited by Tony Hey, Stewart Tansley and Kristin Tolle. Microsoft Research.

89

CLOUD science literature is online, all of the science data is online, and they interoperate with each other. Sadly in 2007 Jim disappeared in his boat off of the coast of California. While his future contributions will be missed – he has certainly put yet another stake in the ground for what might be. HP Story HP has formed a joint cloud computing effort called the Cloud Computing Test Bed (CCTB) that will allow researchers from allied academic institutions and industry to test software written for cloud infrastructure. The idea is to provide a large cloud-computing infrastructure where people can test distributed software at all levels of the stack, from low-level OS, storage, and networking technologies all the way up to user-facing applications. The CCTB will consist of six sites, each of which will house about 1,000 to 4,000 processor cores. When those six initial centers are ganged together into one large cloud, the resulting machine will have a core count that should put it somewhere in the Top 20 supercomputers list. This compute and storage cloud uses HP hardware and software and runs Yahoo's Apache Hadoop, an open source platform that lets users write and run applications that process vast amounts of data.

90

COMPUTE & STORAGE Today’s Clouds are Small Google’s cloud contains over 500,000 servers and many see that as large – but is it really? Just to put this into perspective consider a few other massive computing systems. The NYSE stock trading system processes about 10M trading transactions per day. Visa is believed to process around 100M credit card authorization transactions and Google about 200M searches per day. Now let’s imagine that every mobile phone in the world executed on transaction per day that would be over 4,000M transactions per day. In fact, IBM projects that by 2011, the world will be ten times more instrumented then it was in 2006 and the number of Internet connected devices will grow from 500M to 1 Trillion. Indeed when you now see the iPhone and Android as merely the “eyes and ears” of the cloud to quote Vic Gundotra of Google, then the potential number of transactions will only grow. Consider real time translation, turn-by-turn navigation or determining the supply chain that brought you the lettuce in your local market and you’ll quickly realize today’s clouds are small. Computer Carbon Footprint As compute & storage infrastructure grows and as our concern for global warming increases it will turn out that for modern technology-powered companies the carbon footprint of 91

CLOUD the company’s computers is the new smokestack of the 21st Century. In this section we discuss a simple way to calculate the Computer Carbon Footprint™ (CCF) of any company. The CCF is computed as follows:
CCF =#Employees × SPE× TBY × PUE avg × CPFavg

Where SPE = Servers Per Employee TBY = Technology Birth Year factor X 400 watts
PUE avg CPFavg
= Average Power Utilization Efficiency of the data centers = Average Carbon Power Footprint of the sub-region grids (lbs/watt-year)

Servers Per Employee (SPE) The number of Servers per Employee may be a simple measure of how much a company is powered by technology. Interestingly, in a presentation16 by Simon Commander of the London Business School and Rupert Harrison of the IFS School of Finance they measured the number servers per employee in Brazil at 0.04 and in India at 0.02. In a book17 published by the World Bank in 2007 the on the economy of the country of

16 ICT adoption in developing countries: Firm-level evidence from Brazil and India; Simon Commander (London Business School) Rupert Harrison (IFS School of Finance 1st June 2006. 17 Environmental Priorities and Poverty Reduction: A Country Environmental Analysis for Colombia (Directions in Development) (Paperback) by Kulsum Ahmed (Author), Kulsum Ahmed; Ernesto Sanchez-triana; Yewande Awe (Editor)

92

Koomey. John and Hansell. This is done by establishing a base line of power consumption at 2006 at 400 watts and then if the servers are on average older or younger. The New York Times. Of course probably nothing compares to the modern largescale Internet companies.33 45 India Brazil Columbia Google Figure 4. We base our analysis on a recent paper19 authored by Jonathan G. increasing or decreasing the power consumption.000 people in 2006 it would put their SPE (Servers per Employee) at 45.2: Servers Per Employee Technology Birth Year (TBY) TBY is designed to compute the approximate power consumption of all of the servers.COMPUTE & STORAGE Columbia the range was reported to go from 0.32-3.000. 2009 18 93 .18 With Google employing approximately 10.32 to 3.02 0.04 0. In 2006 the number of servers Google deployed was estimated to be 450.33 computers per employee. Hiding in Plain Sight: Google Seeks an Expansion of Power. 2009 Released on the web: August 17. of Lawrence Markoff. 19 Assessing Trends in the Electrical Efficiency of Computation over Time. Year 2007 2007 2007 2006 Servers Per Employee 0. 2006. Saul. Submitted to IEEE Annals of the History of Computing: August 5. June 14.

doubling every 1. Henry Wong from Intel Corporation. As shown in the figure below. a pace of change in computational efficiency comparable to that from 1946 to the present. The paper explores the relationship between the performance of computers and the electricity needed to deliver that performance.5 years then the same unit of performance would require 4x more power 3 years earlier in 2003 and ¼ the amount of power 3 years later.5 years. in 2009. Figure 4. Technology Birth Year 94 . Marla Sanchez with Stanford University and. computations per kWh grew about as fast as performance for desktop computers starting in 1981.3: Computation per kWH per decade Under the above assumption that computations per server doubling every 1. Stephen Berard from Microsoft.CLOUD Berkeley National Labs.

com/Downloads/svrpwrusecompletefinal.4: Technology Birth Year factor While every server consumes a different amount of power we will use 400 watts as the average power consumption of a server in 2006. RighNow.amd.5: Weighted average power (Watts) of top 6 servers. Class Volume Mid-range High-end 2000 183 423 4874 2003 214 522 5815 2005 218 638 12682 Figure 4. the introduction of compute and storage cloud services has the 20 Koomey. G. J. many companies focused on reducing the cost and as a result gave rise to the first generation of application cloud service companies (Salesforce..COMPUTE & STORAGE 2003 2004 2005 2006 2007 2008 2009 4 3 2 1 (Baseline) 0.25 Figure 4. 2007.). More precise calculations would include the data in Figure 4. February 2007. Taleo. While these companies have wide applicability. Estimating Total Power Consumption by Servers in the US and the World. This will also make our estimates conservative. by sales20 Summary Post the crash of 2001... http://enterprise..5.75 0.5 0.pdf 95 .

CLOUD potential to make an even bigger impact. Again spurred by the economic downturn of 2008 and the tightening of credit anyone looking at deploying new hardware resources (or even upgrading existing infrastructure) has to at least consider this new model of computing. And just as the advent of low cost PCs spurred the last generation of client-server applications, this new generation of compute and storage cloud services promises again to bring a new class of applications we’ve never seen before. Which brings us to the subject of the next chapter – platform services for new application development.

96

PLATFORM SERVICES for SOFTWARE DEVELOPMENT
With the increased availability of compute and storage cloud services just as the battle for the desktop was won by winning the developer, so it may be in the fight to win the cloud. This chapter is not written for the VP of R&D at a software company, but instead is meant to give everyone else, who is not an expert a working knowledge of the issues, tools and economics of software development. We’ll cover the fundamental parts of the software development lifecycle and touch on some of the new horizontal and vertical platform cloud services which are providing software developers with a set of tools tuned for the 97

CLOUD world of cloud computing. We’ll see how new technologies are bridging the worlds of developing new applications, and managing them, which heretofore have been highly separate disciplines. Application Spectrum Before we launch into discussing some of the key success factors, it’s worthwhile noting all applications are not created equal. While all software can be delivered as a service from the cloud, how it is delivered will vary according to the software and its usage. One dimension in variation is what we’ll call the application weight.

Figure 5.1 Application Weight

98

PLATFORM for DEVELOPMENT By weight we mean the complexity of the business process. Email is a very lightweight business process, meaning the way each of us uses email is pretty much the same. This is also true of eBay-style auctioning, Web conferencing and Internet search. An example of a heavyweight business process might be discrete or process manufacturing, areas that Oracle and SAP have specialized in. The way Coca-Cola manufactures soda shares little in common with the process Chevron uses for refining petroleum, although both are manufacturing processes. These processes are much more customized. No one has figured out how to make these as customizable as changing your Yahoo home page. As a result, it’s been the case that lightweight business processes have been earlier to move to Model Six or Seven. The implications of application weight are many, first, on the frequency of change. The lighter weight business processes allow for a greater frequency of change. At one end of the spectrum, manufacturing or financial applications deployed in the traditional model have resulted in upgrades occurring once every couple of years, often at great expense. At the other end of the spectrum is eBay makes changes 52 times a year. Another evaluation you’ll need to make when positioning your offering is to map the number of users for your software 99

CLOUD against the frequency of use. This usage spectrum is shown in Figure 5.2.

Figure 5.2 Usage Spectrum

Corporate applications, like SAP Financial Applications, usually have very few users, and, furthermore, those users use the application infrequently. The closer you get to the consumer Internet, the more dramatic the change in both the number of users and the frequency of use and therefore on the degree of specialization and focus for the application. As you think through your product offering, figure out where in these maps you’re going to fit; it will have many implications for your product and delivery options.

100

a Tenant ID column associates every record with the appropriate tenant. This approach does not require rewriting or building any new software and allows you to offer the service to the 100s or 1000s of businesses who have already bought your software. to be deployed as a service in the cloud and gives the customer some. because multiple tenants share the same database tables. Of course. In this approach. of the flexibility they have come to expect in an on-premise deployment. However. The multi-tenancy approach allows you to serve the largest number of tenants per database server. 101 .PLATFORM for DEVELOPMENT Multi-tenancy One area that is hotly debated is that of a single-tenancy versus multi-tenancy model. you only architected for singletenancy. but not all. and you intend to write the application from scratch. a given table can include records from multiple tenants stored in any order. Adopting Model Four or Five allows your software. architected for traditional deployment. if you have a lightweight business process. to ensure that tenants can never access other tenants’ data. If you’re coming from the world of traditional enterprise software. this approach may incur additional development effort in the area of security. you have to consider a multi-tenancy model.

Most of the multi-tenancy service providers run multiple sites. don’t get embroiled in the religious war here— there are numerous tradeoffs to be made. eBay has maintained one code line since they were founded. how is that different than single tenancy? Bottom line. whether you have a lightweight or heavyweight business process. But just when you thought it was a black-and-white decision. Furthermore each of these applications can be on a different revision level. depending on whether you’re starting with an existing application or building from scratch. the multi-tenancy approach forces there to be one code line and. 102 . and it has exponential advantages when you serve millions of users in a highly fluid environment. Of course. and finally the margins and pricing you’ll need to be successful in the market. In the limit. This is often done to do A-B testing of new releases. this is only possible in areas where you have lightweight business processes. as well as a change management technique. vastly simplifies the development and management of the software. therefore. let’s make it more confusing.CLOUD Perhaps more important. if General Motors wanted one of those sites to run only GM auctions.

If you want the highest quality. you can specialize. lowest cost automobile. not a McLaren F1. computers are far better at repetition than people. ultimately.000.3 Standardization If we were to take a paintbrush and walk around to every computer that has deployed an SAP or Oracle enterprise application. Specialization leads to repetition and.3: dramatically reducing the cost and improving the quality of anything starts with standardization. So how does this apply to software? Figure 5. we’d end up coloring each one differently. choose a Toyota.000 for the car and the need for a McLaren mechanic. as each 103 . If something is standardized. you can’t argue its low cost and high quality. While the idea of a hand-built automobile with a production run of fewer than 100 cars certainly has exclusivity going for it.PLATFORM for DEVELOPMENT Underlying this whole debate is a much more fundamental principal that’s illustrated in Figure 5. at $1.

CLOUD installation would be on different server and disk architectures. Over at the other end of the figure. In 104 . databases. Now. all of the states in between represent progressively greater degrees of standardization. each deployment would have a different change management process or disaster recovery process implemented by different groups of people. and ultimately automate the greatest cost you have—the human labor required to manage these systems. software architecture. Furthermore each would be deployed on a different software stack—different versions of operating systems. think about eBay. every box on the right-hand side has a different color. Software Development Lifecycle Few books on software engineering have been as influential as Fred Brooks' Mythical Man Month. at the time. repeat. the largest single software development project ever undertaken. each running the same software. hardware and management processes. With greater degrees of standardization comes the ability to specialize. So if you look at Figure 5. with three sites. Every choice you make in hardware architecture. Beyond that. Much of the book was based on his experiences as the development manager of IBM's OS/360. and middleware. originally published in June 1975. and management processes should be an effort to move closer and closer to the left-hand side.3.

” Many in the software business can testify to their personal experiences that validate his comment. He furthermore estimated that any software project would use a 1/3rd of its scheduled time in requirements. and large-scale integration did for computer hardware. well over budget. Brooks documented what came to be known as Brooks’ Law: “Adding Manpower to a late software project makes it later. and simplicity what electronics. a 1/4th in unit test. No Silver Bullet: Essence and Accidents of Software Engineering. and a 1/4th in system test time before it would be ready for release. IEEE Computer. Fred. you'd spend six months in requirements. April 1987. No inventions that will do for software productivity. he recounts that the project could not be considered a success.PLATFORM for DEVELOPMENT the book. again no real surprise to the old-timers. reliability. 105 . “Not only are there no silver bullets now in view. Brooks later published a landmark paper21 in which he wrote. He writes of flaws in the design and execution of the system and ultimately concludes that the project was late. Assuming a 24-month release cycle. 1/6th in coding. three months writing code. transistors. the very nature of software makes it unlikely that there will be any. We cannot expect ever 21 Brooks. and underperformed against expectations for the first several releases. and the last 12 months testing.

and tested. Maybe this isn’t the silver bullet Dr. each attempt has come up short. unfortunately. Brooks talks about. 106 . but you be the judge. the industry has struggled to find a way to uncover Brooks’ elusive order-of-magnitude improvement in software productivity. Horizontal Platform Services Once you know the kind of application you want to build it remains to consider how you’ll build it – what your coding framework will be.4 you have three large choices: horizontal software. the movement to delivery of software as a service is fundamentally changing the way software is designed. horizontal services. as you will see in this chapter. and vertical services. As shown in Figure 5. program verification. developed. While building and deploying high quality software is certainly no easy task.CLOUD to see twofold gains every two years.” but. object-oriented programming. and Software Engineering’s Capability Maturity Model (CMM) all have been thought to be the “silver bullet. Rapid prototyping.” Throughout the past 30plus years.

. Horizontal platform services offer you less flexibility. Further restrictions in flexibility (e. All of today’s application cloud service providers. use large amounts of horizontal software on compute and storage services they have had to manage.4 Platform Cloud Services Horizontal software refers to a framework where you may use any of the software (open-source or traditional Model 1 software) on a compute and storage cloud service.PLATFORM for DEVELOPMENT Figure 5.g. Which brings us to horizontal services. definition of a particular database schema) result in more vertical platform 107 . who did not have horizontal or vertical platform services to choose from. middleware and languages of your choice. In this model you can choose to use the database. but allow the cloud service provider to manage the platform software thereby reducing cost and complexity.

Figure 5. for greater productivity and lower cost assuming you’re application maps well to these vertical platform services. In order to understand the advantages of these platform services it’s useful to further understand the traditional application environment.CLOUD services in which you’ll further trade off flexibility. Figure 5. the group which has had to port and test it in multiple operating system and database environments and finally the operations management group which has to manage the application after it’s delivered.5 Traditional Application Development & Deployment What this new generation of platform services is attempting to do is to merge these worlds together since now the builder of 108 .5 shows the traditional silos between the group that writes the original application.

Engine Yard is a good example of providing horizontal platform services to the Ruby-on-Rails community. has introduced Azure to provide an easy on ramp for the millions of Microsoft programmers around the world to build next generation application cloud services using many of their familiar tools. More importantly. does give you access to the highly scalability of Google’s compute and storage services. Microsoft. Finally. As an example these platform services provide: • • • • • Maintain a current level of the underlying software Automatically replace a failed server One-button click to replicate production application for testing and staging Automated change management – from code check in to production Add capacity when load thresholds trigger. while it has a number of programming model restrictions. who ruled the client-server programming model. versus a traditional LAMP stack deployment frees you from the operations management hassles like disk space management. security patch management and backup and recovery. 109 . Google’s App Engine.PLATFORM for DEVELOPMENT the software is no different than the people who have to operate the software.

founder and then Chairman of Microsoft urged company leaders to “act quickly and decisively” to move further into the field of offering software as services. dated Oct. The Internet Tidal Wave. Bill Gates. Ray Ozzie. “This coming 'services wave' will be very disruptive. First and foremost is Chief Software Architect. which prompted a massive shift at Microsoft toward Internet-based technology. “We’re going to look back at this era and wonder how we did without this other kind of computer in the cloud.CLOUD Microsoft Azure Story In a widely publicized email to top executives. While Gates might have left day-to-day operational responsibilities.” Gates compares the push toward such services.Net services with 110 .” Windows Azure seeks to provide the Microsoft Visual Studio programmer with compute and storage services similar to those provided by Amazon as well as SQL and .” Gates wrote. to the changes he saw nearly a decade ago. 2005. “We have competitors who will seize on these approaches and challenge us—still. the opportunity to lead is very clear. In a recent interview Ozzie said. which range from online business software offerings to free Web-based email. Then he wrote a now-famous memo. there are key people to carry on his vision. “The next sea change is upon us. 30.” Gates wrote. in order to best formidable competitors.

It's a fluid system that runs your code in response to load and demand. By providing a more vertical (more constrained) offering the tradeoff will hopefully 111 . or RAM or a number of CPUs. Google AppEngine Story In mid-2008 Google introduced App Engine. App Engine’s goal is to provide a simpler alternative to the traditional LAMP stack. App Engine is not a grid computing solution. App Engine Tech Lead said. We also don't give you a raw virtual machine. and then we run and serve it for you.PLATFORM for DEVELOPMENT integrated operations management. specify how you want it to run in response to requests. While with Google Apps they provide office applications in the cloud. We don't run arbitrary compute jobs. or anything like that. “App Engine is different than a lot of other things out there. You don't reserve resources. or machines. As an example the SQL Azure service provides availability monitoring on five minute intervals as well as updating lower level system software as required. App Engine is their first attempt to provide a platform for others to build cloud-based applications.” Ultimately. we provide a way for you to package up your code. Speaking at App Engine Campfire One in April 2008. Kevin Gibbs. Instead.

the system used for Google.CLOUD be greater productivity at the expense of flexibility. which can be created as you code. making all of them appear to be a single storage table.com. since they are typically a source of performance problems in a distributed system. As we know. but there also needs to be an operations environment. App Engine starts by writing server side code in Python. It is designed to move data around and restructure the system automatically to account for hotspots and increased storage. If you want flexibility you can always go buy your own computers. Bigtable is schema-less. distributed system that can span tens of thousands of hard disks on thousands of machines. While there are some early applications being demonstrated. One feature you won’t find is joins. meaning it can support arbitrary new properties or columns. 112 . Unlike SQL. not only does there need to be a programming environment. You can make use of numerous Python modules and libraries using any Web framework. it is still early days. The database is not SQL instead it is based on Bigtable. Toward that end an administration console is provided to give you access to the application. Bigtable is a fault-tolerant.

PLATFORM for DEVELOPMENT Vertical Platform Services Further down the spectrum of decreasing programming model flexibility for increased productivity and decreasing cost to manage are the vertical platform services.com and Facebook’s platform. Netsuite seeks to allow application builders who want to specialize their ERP oriented platform for more specific markets. Salesforce. Salesforce with force. ironically almost at the 113 .com has led the way in turning their platform into a service others can use. by opening up the platform (most notably Facebook Connect) can leverage a wider development community to extend their services far faster than they can with their captive R&D group. AppExchange was launched in late 2005.com seeks to provide not only the technology. Vertical platform services assume you are building a specialized usage of the wider platform provided by the vertical platform cloud service. This typically means you’re leveraging at least the data model and hopefully some of the higher level abstractions provided by that data model. Finally with so many users of Facebook.com’s force. but also a marketplace (AppExchange) to encourage new CRM applications. Good examples of vertical services include Netsuite’s NSBOS.com Story In traditional business software. Salesforce. Salesforce.

The other key event was that developers began to build applications that were not the conventional formsbased apps. Visualforce is a framework to build custom user interfaces.com 114 . according to Phil. enabling productive code reuse and standardization.com platform with a procedural language so that users can go beyond declarative programming to write code where needed. writes that two events changed Salesforce. the objective was to enable customers and partners to customize the Salesforce. You can use this language to create triggers. The result. which required transactional integrity.CLOUD same time that Oracle acquired Siebel. A year after the introduction of Apex. Apex can also create classes. similar to HTML. Phil Wainewright.com platform added Visualforce.com’s approach. was the creation of the Apex programming language ultimately leading to a complete platform for building application cloud services. There was no firm plan to extend the core application platform beyond its formsdriven database model.com application and extend it. Force. Apex provides Force. The first was an attempt to use the platform to write a quote application. Underlying the Force. At the time. which can be called from other classes and triggers. who blogs about SaaS for ZD Net. which fire when any user performs certain data operations. The Visualforce framework includes a tag-based markup language.

launched in 2008. as well as optionally providing user interface representations in both the parent and child records. foreign keys. While similar to SQL in some ways. Instead of developers’ having to deal with keys. which he started in 1994. not joins.com database differs from relational databases in the way record relationships are implemented. the real test is the types of applications that can be built and delivered using the salesforce. Bob Morrell founded Riskonnect in July 2007. Riskonnect’s software. Two interesting case studies are Riskonnect and Stakeware. for a more intuitive navigation of data. “This is 115 .PLATFORM for DEVELOPMENT platform is a multi-tenant database. He’s had a long history in risk management software beginning with Risk Laboratory. The Force.com platform supports two query languages: • • Salesforce Object Search Language (SOSL) is a language for searching all across all persisted objects. Salesforce Object Query Language (SOQL) is a queryonly language. the database deals in terms of relationship fields. primary keys and defining relationships in terms of these keys. The Force. A relationship field stores the ID of the parent record in a relationship. Of course. helps people visualize and understand risk at very high levels. it's an object query language that uses relationships.com cloud.

lecturer.com application.CLOUD essential because too many consultants only use complicated calculations. and Shell Oil. Each of these examples is just further proof of what Marc Benioff said: "Software developers from a developing country can build just as great an application on our platform as somebody who lives in Palo Alto.” Morrell said. An example would be if one factory goes down. Riskonnect allows users to drag and drop information to compare risks visually. another force. The Mexican Government. in 2004. and analyze stakeholder data. identify key issues and related indicators that are relevant and material to include in the standard G3 sustainability report. Natan Zaidenweber founded StakeWare. The software also includes the ability to monitor and analyze emissions to help provide appropriate trade-offs when lowering the carbon footprint as well as the management of sustainability policies across supply-chain. Zaidenweber has over 15 years of experience in sustainability as a researcher.” 116 . But many people don’t need that. how would other factories and the supply chain respond. and manager working in projects for The World Bank. organize. StakeWare enables corporations to gather.

with their true identity. it’ll be difficult to pretend you’re Angelina or Brad. That kind of growth and scale can in part be attributed to Facebook opening up its platform and providing a full suite of APIs—including a network protocol. This is a particularly valuable service since in the cloud many people will not represent. nor are incented to represent themselves. How many times have you signed up for a site or a newsletter with fake identity? On the other hand it’s difficult to have a fake Facebook identity. a little more than a month after announcing it had 150 million active Facebookers. Their most recent effort has been to bring the power of Facebook to other applications by providing the ability to: • • • Connect any existing Facebook account and information with any application. twittered the number was now 175 million. Connect and find friends who also use the application. Of course not only is basic profile 117 .PLATFORM for DEVELOPMENT Facebook Story In February 2009. Share information and actions on the application with their Facebook friends. and a text markup language—that allow applications to integrate tightly with the Facebook user experience. after all unless you have no friends. Dave Morin. who runs the application platform team. a database query language.

If you’re interested in seeing early examples of applications using the Facebook platform take a look at TechCrunch. groups. their privacy settings will follow them. With access to the social graph. • Social – Helps people interact and communicate more effectively by using information from the social graph 118 . VLane. Some of these are clearly unique to their applications—others are just good ideas. an application can dynamically show which Facebook friends already have accounts on the application. friends. and more.CLOUD information available but also the profile picture. For example. Citysearch. events. if a person changes their profile picture. and Howcast. Interestingly Facebook has published a set of principles for their applications. photos. this is automatically updated in the application. the individual can control who can see what pieces of their information—the same rules that they set on Facebook can be applied through your application. ensuring that users' information and privacy rules are always up to date. or removes a friend connection. As an individual moves around the Web. In addition. Applications should be meaningful.

• • • Clean – Designed to be intuitive. • • • Secure – Protects user data and honors privacy choices for everyone across the social graph Respectful – Values individual attention and honors their intentions in communications and actions Transparent – Explains how features will work and how they won't work. 119 . When iLike launched 10.PLATFORM for DEVELOPMENT • • • Useful – Delivers value to people by addressing real world needs. from entertainment to practical tasks Expressive – Enables people to share more about who they are and about the world around them. easy to use and free of mistakes Fast – Achieves low latency while scaled to handle individual demand Robust – Maintains reliable uptime and minimizes error rates With such a large population of people on Facebook. Engaging – Provides a deep experience that people want to come back to regularly Applications should be trustworthy. a successful application on Facebook can grow to millions of people in a short time. A case in point is the story of iLike. especially in triggering person-toperson communications Applications should be well designed.000 people joined in the first 12 hours.

we can now economically reach the Fortune 5. 120 . a mechanism to distribute applications. The resulting specialized application and data reside inside the NetSuite platform and share database and server resources within the core solution. and the NetSuite application itself. NSBOS targets building out the core NetSuite offering with additional specialized functionality beyond the services delivered by the core application. In less than a year. Now if we all were so lucky.000.000". NetSuite Story In 2005. As a more vertically focused cloud service.CLOUD By the end of the first day they had 50. a public cloud development environment that includes a set of Application Program Interfaces (APIs).000 per day. To use Zach Nelson’s words: “By using the cloud. iLike announced they had passed three million and were growing at a rate of 300. This tight integration enables the platform and applications to remain in sync as the underlying NetSuite application is upgraded.000 people. NetSuite opened its architecture to partners and outside developers to further its reach through its NetSuite Business Operating System (NSBOS).

Java or PHP. SuiteTalk uses industry standard SOAP and XML messaging to communicate with external applications.com.Net. the JavaScript API. and create completely new custom record types for objects unique to a specialized application. SuiteScript. allows developers to add workflow to records and automate business processes for target industries with full transactional integrity. founded Rootstock. give application developers data storage flexibility. Pat Garrehy. The platform also allows developers to build Web services integrations using . “Rather than cast a wide net. As a result of this strategy NetSuite has focused on specialized developers. NetSuite VP of Developer Programs.” says Guido Haarmans. Developers can create custom forms to give users access to specific fields. Finally. a veteran of the manufacturing and Material Requirements Planning (MRP) software. A couple of examples of these specialized applications are those provided by Roostock Software and SPS Commerce. Integration with external systems is supported through SuiteTalk.PLATFORM for DEVELOPMENT The NSBOS APIs. called SuiteFlex. we target partners with the industry experience to create compelling applications. the resulting application can be packaged using SuiteBundler. which enables members of the SuiteCloud Developer Network to distribute their applications on SuiteApp. His previous company developed conventional 121 .

they built a multi-plant MRP application. most of software development time and cost involves testing. Testing As Fred Brooks noted. But without an EDI solution they were limited in which retailers they could work with. benefited. from inventory to shipping and invoicing. He recognized the rise of cloud computing and rather than replicate fundamental ERP functionality he chose to build on NetSuite. SPS Commerce. thereby allowing a small supplier to quickly and inexpensively meet the complex EDI requirements. Circle of Friends and Mad Croc. in just nine months. a leader in electronic data interchange (EDI) supply chain developed a specialized application that allows smaller suppliers to tap into their network of over 1. So how does moving to an on-demand 122 . Garrehy got the itch to return to the MRP business. Neither could afford traditional ERP and EDI applications and the IT resources necessary to keep them running.300 retailers. both small CPG manufacturers. With NetSuite providing the underlying cloud services. By adopting the vertical application cloud service they could quickly and inexpensively manage their entire trading relationships. In 2008.CLOUD Model One ERP/MRP software that was used by companies like Lockheed and Solectron.

Whether that’s a standardized delivery model in Model Four or Five. Furthermore. or using one line of code when you move to Models Six or Seven. the QA team can limit their focus to one line of code (hopefully). you can use the actual customer systems to develop real-world tests. since the application will have already been tested in the development labs. Now. instead of trying to create and manage tests for this application release (and the three previous releases) and insure that they work on Oracle. The tests are more representative and come at a much lower cost. 123 . you’ll know. moving to the model reduces the number of variations and creates greater standardization. Rather than employ an army of test developers in your engineering organization. Collapsing the supply chain between the customer and the producers of the software has even further economic benefits.PLATFORM for DEVELOPMENT model change anything as regards testing and QA? First of all. IBM and MySQL databases on HP. IBM or Sun hardware. one database. So. reducing the degree of freedom reduces the number of overall tests that have to be developed and executed. the model lets you begin to leverage tests of specific customer configurations and execute those tests very early in the development cycle. rather than wonder whether an upgrade will work. and one hardware platform.

just as we will see in the next chapter on operations. Concur co-founder Michael Hilton said. not a release every 24 months. that means over six thousand companies and over four million end users are affected by any defect. That means that. “This fundamentally changes how you develop. our users are getting new innovations and benefits without having to do anything. And you end up releasing more often. 100 percent of your customers experience that bug immediately.” It has fundamental impacts on how you think about quality assurance.CLOUD But there are also major new responsibilities and issues. You end up releasing things in smaller chunks. “For Concur. Think about that.” Of course. so users can easily digest them. But there’s a flip side as well—you can fix the issue and 100 percent of your customers have that issue resolved—no more waiting for the slow trickle of customers to download and install the patch. Lots of things end up being optional. It forces you to care a lot more about the user experience. meaning 124 . At Concur. we do meaningful functional updates to our service every single month. if you introduce a bug into production. ease of use and change management. For instance. the testing/QA function will need to re-engineer for the new reality of a release a week or a release a month. Again. as Hilton points out. every month.

today Senior VP of R&D at Risk Management Solutions. “You can have any color you want as long as it’s black.” Customization One of the major objections to the SaaS/application cloud service model has been the belief that there is no ability to customize. Marian Szefler. Some engineers would try to be PC (politically correct) and say. it remained challenged on how to address this question. While this subject is almost nonexistent from a consumer perspective. As Oracle On Demand was blessed with a natural channel to medium and large-scale businesses. architected an innovative approach while at Oracle that changed the argument from techno-babble to a business decision. any conversation with corporate IT will inevitably lead to the discussion of how to customize the application.” The conversation was confusing.” was the famous Henry Ford quote often used by people describing early offerings.” Or others would say. “We don’t support extensions.PLATFORM for DEVELOPMENT that. if the feature is big enough. we allow for configuration. 125 . customers don’t have to deal with it right away if they don’t want to. we support configurations. “We don’t allow for customization.

CEMLI stood for Configuration. “CEMLIs” could be counted. grouped.6 CEMLI Framework The framework shown in Figure 5.6 lists the 17 classes and is. one could have a precise conversation. Extension. You should take away a few ideas from this. Localization and Integration. of course. unique to Oracle applications. by creating these classes.CLOUD In 2000. and described 17 classes of reason why someone other than Oracle R&D might need to add software to an Oracle application. Figure 5. Modification. for the first time. First. and discussed by both the customer 126 . Marian defined the CEMLI framework.

So now the CIO could ask the business. something you want to avoid at all costs.PLATFORM for DEVELOPMENT and Oracle—the framework eliminated the politically correct dialogue that preceded it. the customer could choose how much they wanted to spend. Second. for what degree of uniqueness. the CEMLI framework created an environment where the degree of customization became a business decision. Adding a new report was far less invasive than adding a trigger. Fred Magner. In fact. not all CEMLIs were created equal. “Do you want to spend X for the uniqueness. where standard products. The classes were ranked by how “toxic” they were. used the cost to manage CEMLIs to move the number of CEMLIs down from 1500 to 300 over the period of 18 months. you were coming closer and closer to making fundamental modifications to the software. The model created a framework for the business to justify why the particular CEMLI was worth the extra expense. By isolating CEMLIs and separately pricing the management of these. not a technical or religious one. meaning the likelihood that a mistake in this class would cause data corruption or a down system. former CIO of Unocal. Finally. There were cases at Oracle where the cost to manage the CEMLIs was larger than managing the entire Oracle application. for example. like Toyota 127 . or can we use the standard product and save money?” This is no different than in the real world. as you got to higher degrees of “toxicity”.

are far lower cost (and higher reliability) than custom crafted cars like a McLaren F1. When you think about traditional enterprise software companies. a lot of times the teams that write 128 . Taleo CEO Michael Gregoire said. years later. However. The cost for servicing the “one” is many more times than the cost of servicing the “many. when some of these very same prospects have returned to look at Taleo to replace those same customized solutions purchased earlier from other vendors. Taleo has addressed the customization challenge by creating a single line of code for all customers while baking in the ability for customers to do extreme configurations. Concur created a model that allows them to easily move customizations into core code.CLOUD Corollas. This approach provides customers the ability to avoid the customization version lock trap while still enabling customers to meet their unique business requirements. our approach has been validated.” Concur’s Michael Hilton sees the investment in architecting a solution as one reason traditional R&D and consulting are now connected. He said. “It’s just part of our development model. “This approach did result in us having to pass on some very large opportunities in our formative years when prospects insisted on a customized approach.” Anyone in enterprise software has faced these challenges.

utilizing the same tools and processes. we built out a fairly complex architecture to deal with customizations. followed by ten years with NetSuite. Both teams need to be working in the same organization. it is much easier to talk to the customers.com takes special advantage of 129 . Perhaps no one is better qualified to talk about the differences between the traditional development and development in Model Six than Evan Goldberg. having spent eight years with Oracle.PLATFORM for DEVELOPMENT customizations (typically consultants in the field) are completely separate from the core R&D teams. It took us a couple of years to really get it right. migrates. as Founder. He speaks from an excellent vantage point. and CTO. At Concur. Amazon. In our model. First. What’s more.” Traditional vs. In a lecture at Stanford University. let’s get back to R&D. so that everything scales. Our large clients can have their unique needs meet. you can tell what features are being used. that doesn’t work. there is no “middleman” between the producer of the software and the consumer. and works together seamlessly. what features aren’t. Chairman. Goldberg pointed out there are numerous product design benefits to Model Six. but the benefits are extraordinary. Cloud Development But. while still operating in our multi-tenant environment. As you’re delivering the service. since the software can be instrumented.

because there can be universal adoption of innovations. delivering software as a cloud service allows for testing through 130 . They can use data to determine what features to keep and what to retire. compare with the old system. you can build on top of previous innovations rather than worry about which customer has which features. “they were never good at. something. given that installation is occurring on the companies’ computers. Goldberg said. versus working on back porting to multiple releases on multiple platforms. “Kill Bugs Dead!” There is instant gratification to solving complex problems. as Goldberg put it. versus the traditional model. the engineering staff doesn’t need to spend a lot of time on simplifying installation. As we’ve already pointed out. software engineers can put all of their energies into that release. Finally. The immediate feedback loop afforded by the model allows the software engineers to.CLOUD designing an operational environment that allows them to experiment with new features and. where it may be months before a particular patch can make it into a production system. And. With an ability to focus on single version. at the same time.” The last group Goldberg addressed in his Stanford lecture was the QA organization. as a designer.

With such a rapid rate of change. we can’t depend solely on human power so all aspects of engineering are going to have to automate.” He noted that.PLATFORM for DEVELOPMENT actual customer usage. for a VP of R&D in a traditional software company. Goldberg ended his comments by putting up a slide that said: “Operations: Welcome to the New World. The challenge that QA shares with the other disciplines is that each release of software is happening much more quickly. But in any of Models Four through Seven. There was always the support organization and the customers own operational people that stood between you and any issue. In the case of NetSuite. worrying about the operations of your software was a distant issue. the average has been over three releases per year. Whether it’s database architecture. how you architect for 131 . Summary We’ve attempted to highlight some of the more important differences between traditional software development and the new world of developing software for cloud deployment. The ability to monitor real-time performance and develop and inventory tests based on actual usage are significant advantages over the traditional model. that idea is a distant memory.

you’re going to need to understand how to manage the security. change and. perhaps most importantly. it’s a new world. we’ll spend time on the key success factors in the management and operations of software.” 132 . the availability of your software. or how you’re going to have to re-think testing and software release. As Evan says.CLOUD customization. problems. In our next chapter. performance. “As the VP of R&D.

This is no different in the world of managing software applications. an application cloud service provider manages hundreds of systems and therefore has to repeat numerous processes multiple times. vice president for Oracle.PLATFORM SERVICES for OPERATIONS Repeat after me. This is true for all human endeavors. cites an example of where this repetition ultimately 133 . the key to excellence in anything is specialization and repetition. and a great surgeon performs hundreds of surgeries to master her craft. Today. Don Haig. a student recites multiplication tables to learn to multiply. Swimmers swim thousands of laps to perfect a stroke.

they will do this upgrade once and only once. can often separate the talented from the average: How often do you do this? Research has established that busy hospitals generally deliver better care. said. an event that most organizations have to manage with some frequency and yet is an annoying distraction from their core business. but in most IT departments today. but understand they need to stay current on the latest release of a vendor's software. But recent work adds more detail. upgrades have been repeated hundreds of times per year.” Yes. “Take software updates. CIO. the research shows. TRICARE Military Health Services. who was trained as a neurosurgeon. Consider the following research in the medical profession in light of the complexities of managing IT infrastructure. complex surgeries are done for the first time by people who read the book the night before. A simple question posed to surgeons. Navy Captain Brian Kelly. “You don't need to explain this further.” Upon hearing this example. the users of Oracle applications. CIOs really don't like to do updates. We have an expression in my business—‘You don't want to be the first guy to get your cabbage worked on’. When a customer is managing their own systems. In Oracle On Demand.CLOUD helped his customers. citing the 134 .

” Do we ask the same of our operations staff? How many times have the operations staff upgraded from Release 1 to Release 2? Some surgeons are convinced that measuring patient volume alone is too crude a measure for grading surgeons. Since it may be the first time they've seen the issue. “It is now a very legitimate question to ask a surgeon: How much experience with a procedure do you have? What are your complication rates?” Tufts Medical Center's surgeon-in-chief. the more I fall into a standard routine. who don't frequently and regularly perform certain procedures. they don't know how to react. the profession sees constant repetition as key to mastering complex surgeries. William Mackey said. Dr. a delicate procedure to clear a key artery of blockages.” Many studies of recovery from computer outages show the outage is often extended because the operator has seen something out of the ordinary.” said Mackey. if I encounter something out of the ordinary. I'm quicker to recognize it and correct it. “The more often I do a procedure. a vascular surgeon who performs frequent carotid endarterectomies.PLATFORM for OPERATIONS specific risks patients face in the hands of low-volume surgeons. “Then. ”It’s something every surgeon should keep track of and show patients if requested. But increasingly. 135 .

CLOUD So what’s a high degree of repetition? Dr. If you went bowling ten years ago. If we know the performance management process. Consider a corporation's system. So. then why not program computers to automate the process? To better understand what it means to automate key processes consider a simpler problem. since scoring was left to whoever said they remembered 136 . rest assured. Mark S. How many times a week have they performed a particular procedure? Is the procedure even written down? Can it be repeated? The good news is that if a process can be repeated 100s of times it can be automated. or the upgrade process. after “Where do I get a beer?” was likely. and we’ve repeated it a hundred times. “Who’s going to keep score? Who knows how to score?” You can be certain the same conversation was occurring in all ten bowling lanes. authored a study where high volume meant 40 or more procedures annually. and database and application operations teams. And computers are far better and lower cost than people. a University of California at Los Angeles urology professor. Litwin. each with a different answer to the question. did a score of 200 on Lane 1 compare to a 180 on Lane 2? It was anyone’s guess. network. the first question. while all ten lanes were bowling.

It might not improve your bowling. Ten years later. versus putting it on the shelf and avoiding the upgrade at all cost. they get to receive the benefit of new R&D. the story is quite different. You can be guaranteed that a score of 200 in Lane 2 and a score of 180 in Lane 3 are scored identically. performance. availability. this is good. or VP of R&D will need to focus on. you enter your name. Now. In the remainder of this chapter. Consistency and standardization are being enforced through automation. And from the customer’s point of view. but you can make sure that the scoring process is being implemented consistently. Change Management One of the advantages of Models Four through Seven is the increased simplicity of moving customers forward to the latest release. and the computers score and tell you whether you get to bowl again. we’ll go through some key processes in security. if you go into a high-tech bowling alley. since your developers can focus on one code thread versus being forced to pay attention to multiple threads and platforms. 137 . From an R&D point of view.PLATFORM for OPERATIONS how. or play Wii Bowling. and problem and change management the VP of Operations.

Limit the locus of change 2. 2001. Architect the software for change (roll-in & roll-back) 4.CLOUD We should also note that upgrades or changes to enterprise business software have always been an expensive and difficult proposition. ERP Trends. The Conference Board Report. according to AMR Research.. and Cooke.P.J. Two Northern Illinois professors documented22 the cost and challenges of upgrading Enterprise Resource Planning (ERP) applications. Number 3. with costs that were on average 25 percent over their original budgeted amount23. Communications of the ACM. Gelman. L. W. D. Instrument the environment. 23 Peterson. 22 ERP II: best practices for successfully implementing an ERP upgrade. Change management for applications. Consider these four fundamental principles when designing your process: 1. Factors that range from weak project leadership to minimal employee product training have resulted in many ERP projects being delivered late and over budget. 138 . Volume 49. pages 105-109. The cost of a typical ERP implementation in a Fortune 500 company was estimated as between $40M and $240M.. platform and system software sits at the nexus of the cost and reliability of delivering applications as a cloud service – there may be no more important process. 2006. Standardize the frequency of change 3.

eBay Story If information moves along an information superhighway. instead componentize the application so you can deploy on lots of little boxes. A pool is a set of computers dedicated to a particular function. As the Senior VP of Product. then Lynn Reedy is a master traffic engineer. user interface design. architecture and release management. As you are able to improve the predictability and improve the speed you’ll not only reduce cost. but also increase the reliability. usability testing. make sure you’re not dependent on one large database but find a way to segment and componentize. product specification. Development and Architecture at eBay. Reedy led the group that was responsible for all of the initiatives that pertained to the company’s work in software development. don’t engineer for one big box. 139 .PLATFORM for OPERATIONS And for the key process From-Software–Change-InitiatedUntil-Completed measure the duration in minutes. she says to insure a high availability application. First. Reedy has some good counsel for anyone building software that is delivered on demand. Furthermore. eBay architected a system that was based on the idea that 100 percent of the code base was sent to all of the pools of resources.

eBay architected the software so that code could not only be added.000 regression tests. with new software headed out on the next train. it could also be rolled back. there is a “sell your item” pool and a “My eBay” pool. Even within a pool. The pooling architecture allowed change to be made gradually. As a result. As a result. when software development was done for a new feature. thereby. a new feature would be added to one box. Once it was scheduled for a train (along with other features). the new threads were forced to merge and. For large projects. you need to do this while maintaining availability of the service. referred to as the eBay train. This put the impetus on the new development to resolve problems or conflicts with features coming from other organizations. QA tested the feature for two weeks. when you make a change. eBay’s change management process. all database changes resulted in fields being added.CLOUD For example. pick up the changes every two or three weeks. 140 . Software needed to be both forward and backward compatible. then 25 percent of the pool and finally 100 percent of the pool. eBay architected the system for change. And remember. With the eBay Train approach. the feature was released to QA. the feature was subject to a battery of over 100.

PLATFORM for OPERATIONS

eBay knew that, just depending on the quality of people and process for high productivity and reliability would not be enough. As a result, the company invested in numerous pieces of infrastructure software. At the top of the list was the central application logging (CAL) facility. CAL is both software and an engineering practice to fully instrument the application. Any exception within the system would show up as an error on the CAL graphical user interface (GUI). These components would be seen as a “red pool.” Exceptions could be in the code, network, or data layer. Before CAL, eBay personnel took weeks to understand why a component “threw an exception.” With CAL, resolution took minutes. eBay’s second major investment was in automated rollout and rollback. With over 200 features being worked on at any one time, and with an average of 15 features per train, eBay’s usage of automated rollout and rollback were critical. Running a system that requires 7x24x365 availability, even in the event of earthquakes, tornadoes, or hurricanes is a day-one requirement. As a result eBay maintains at least three data centers in the United States. Some people have deployed fail-over sites that run “dark” and wonder, in the event of a failure, if the systems and processes could handle the failover. Instead, eBay runs

141

CLOUD every site hot and limits each site to no more than two-thirds of the overall traffic. Reedy offers three things that every VP of Operations and R&D should focus on: right culture, right processes, and right metrics. Right culture means a culture of people who find a way to “take the hill,” who think of ways to succeed, not fail. Right process means having a daily, repeatable process that everyone can do over and over again—what works for surgeons should work for IT. And finally, right metrics. Metrics tell you where you are and where you want to be. Making this as obvious to the operations team as earnings per share was to the business unit led to a standard of excellence—few can argue with Reedy’s success. Availability Management Most people consider the most reliable, available system in the world to be the U.S. telephone system. We often refer to “Dial-tone” reliability to describe the highest level of availability. On April 6, 1992, the FCC required telephone service providers to begin reporting outages of 30 minutes or more, which potentially affected 50,000 customers. Recently the standard has been raised to both report on outages affecting 30,000 customers as well as to report on any outages affecting

142

PLATFORM for OPERATIONS major airports, 911, nuclear power plants, major military installations and key government facilities, regardless of the number of customers/lines affected. The results may surprise you. For the past 10 years the data has been collected. These are the summary statistics.

Figure 6.1 Outage Profile

143

CLOUD These outages were caused by everything from Heisenbugs to human error to heavy rain. In Jim Gray’s landmark paper24 he analyzed a set of Tandem NonStop systems and determined that the reasons why systems stopped fell into four large categories, which operator/human error being the majority. In a more recent study25 of more than 500 component failures and dozens of uservisible failures in three large-scale Internet services, it was observed that (1) operator error was the leading cause of failure in two of the three services studied, (2) operator error was the largest contributor to time to repair in two of the three services, and (3) configuration errors were the largest category of operator errors. Operators may face such difficulties because computer designers and programmers have frequently sacrificed ease of use in the quest for better performance. Database software, for example, can require a full-time staff of trained administrators to manage it. Ironically, because hardware and software have grown cheaper over time, operator salaries are now often the biggest expense in running complex Internet sites. The experience in the field leads to forming a handful of fundamental principles in availability management:
24

Why do computers stop, and what can be done about it? Tandem Technical Report TR-85.7 www.hpl.hp.com/techreports/tandem/TR-85.7.pdf 25 Why do Internet services fail, and what can be done about it? David Oppenheimer, Archana Ganapathi, and David A. Patterson; University of California at Berkeley, EECS Computer Science Division, Berkeley, CA, http://roc.cs.berkeley.edu/papers/usits03.pdf

144

PLATFORM for OPERATIONS

1. Develop a high speed, predictable recovery process: problems are going to happen, so design systems that recover quickly. 2. Instrument to pinpoint the sources of faults 3. Build systems that support an “undo” function, so operators can correct their mistakes. 4. Standardize the key management processes and take operations people out of the loop In the end the key availability management process is FromThe-Time-The-Failure-Is-Detected-Until-The-Time-The Application-Is-Live is perhaps the most important process. Furthermore as the recovery time is made predictable and reduced not only will you have happier customers, but the cost of providing high reliability will go down. Performance Management Your customers expect you to provide them with application access at an acceptable level of performance. Performance and availability management are related. If you expect that result in 2 seconds and it takes 200 seconds does it really matter if the application is slow or down. Performance management starts with the difficult task of workload characterization. Every application has some key queries, transaction flows, searches, 145

For many. Once identified these workload characteristics can be monitored. While a discussion about securing online applications could take the remainder of the book. A good example of a company totally dedicated to this is Keynote that allows you to monitor the performance of your web application from points around the globe. which characterize that application. the push to put information in the cloud raises the specter of information breaches. perhaps more important. providers will have to convince IT managers and CIOs that the services they offer are reliable and. Security Management Before enterprises can reap the benefits of on-demand software. secure.CLOUD etc. Perhaps the most famous one you see every day is in the upper right hand corner of every Google Query. we’re going to highlight a couple of 146 . Of course once there is evidence of a slow application then we’re back to using the change management processes we discussed earlier in the chapter. Now detecting a slow application is an exercise in profiling the performance of the application and then seeing when the application is performing outside of those norms.

g. Encyrption also solves problems of protection when data leaves one application domain and enters another (e. • Authorization. The simpler the definition of the security policy and the clear evidence that it is implemented insure the highest degree of trust. Audit considerations include the process to access audit logs. PIN encryption). 147 . Whether thru simple password management. network) encryption becomes the service of choice.PLATFORM for OPERATIONS specific areas you’ll need to think about to insure the applications are secure.. • Encryption. Authorization mechanisms ensure authentication individuals have access to authorized data per a security policy. removable disk. At the core of all security management is auditing. on a tape. retention durations and evidence of tamper proofing.g. all the way up to two-factor authentication with retinal eye scan all security mechanisms begin with knowing who you are – authentically. When physical data protection offered by the co-location facility is not possible (e.. • Authentication. • Audit.

vice president at the 200-person company. a medical device manufacturer. One should be mindful that a SAS 70 report only documents the internal control practices of an organization.CLOUD In the security management area there are numerous standards. without offering any judgment as to whether they are satisfactory. 70 (SAS 70). data security. and so on. as devices are prescribed by doctors and dispensed to patients. A SAS 70 audit is performed by an independent auditor and results in a SAS 70 report. but that you examine it thoroughly to determine whether the provider is able to comply with your own internal standards for privacy.com’s software to monitor the activities of its salespeople and to track its entire inventory. Care Rehab audited Salesforce. One of the most recognized commercial standards is the Statement on Auditing Standards No.” said Ed Barrett. security was an important consideration when the company was evaluating Salesforce. The company.com. Care Rehab Story “For Care Rehab. which makes traction and electrotherapy devices used by physical therapy clinics.com’s security practices before agreeing to 148 . Due diligence therefore requires that you not only request an SAS 70 report from a prospective provider. has been using Salesforce.

one of the largest banks in Singapore.com’s security practices. “Their security is superior to what we provide for ourselves. of course.com.PLATFORM for OPERATIONS use the software. In my introductory remarks. I made a point to tell the Secretary we could implement the Oracle On Demand model either in an Oracle data center in Austin. you have to have the best people in security and the best redundancies. or a data center of his choice. said yes.” I. I had the opportunity to speak with the Secretary of Treasury of Mexico. I'm sure we aren’t the world’s best security people. But I ended my answer by sharing a conversation I’d had the previous week with the CIO of OCBC Bank. “Tell me about your security. but asked him 149 .” said Barrett. “If you’re Salesforce. He said.” Mexico Story We’ll end this section with one personal story. In that situation. mid-way into my explaining the breadth of on demand services the CIO said.com staff members showing Care Rehab how they secured the data that was stored on their servers and reading documents describing Salesforce.” I. Texas. through encrypted VPNs. I spoke for about ten minutes and then asked him if he had any questions. through SAS 70 Level II audits. of course. A few years ago. went through all the security features from the data center. [We] need to have the best salespeople. That audit included Salesforce. “I want you to take my HR and Payroll data. presumably in Mexico City.

since security professionals agree that most IT-related attacks arise within the organization. the challenge of customer service management is not Bruck. Survey Reveals Scandal of Snooping IT Staff. 2007.CLOUD why was he so adamant. four-fifths of all IT-related attacks arise within the organization. with one-third stating a belief that they could do likewise with little risk of detection. I know your folks can see the same information. May 30. their bosses’ salaries and their bosses’ bosses’ salaries.com. with any software that has a reasonable degree of maturity. 2007. which found one-third of them admitting to abusing their administrative access privileges to examine sensitive data unrelated to their job functions. Press release. According to security professional Michael Bruck26. Security Threats from Within. He replied. Bruck’s estimate is consistent with a survey27 of more than 200 IT professionals. About one-fourth reported knowledge of former employees who retained access to sensitive networks long after their termination. Cyber-Ark Software Inc. but what would it mean to them?” It’s a perfect story to end with. Entrepreneur. June 28. Michael. 27 26 150 . Customer Service While many think of customer service management as defect management. “I know my DBAs can see their salaries.

personal to you. Just one-tenth of one percent (0. known information. in a post-Google era haven’t you wondered why was the information so difficult to find? Search-based Applications Today's surface Web (which Google. 10. purchasing. Sales. So if you’ve ever been on hold with a call center. The simple math says the deep Web is. But this surface Web of less than 100 terabytes pales in comparison to the deep Web of information contained in every business. The analysis was quite the opposite.000 times the size of the surface Web. HR.000 terabytes) of hard disk capacity was shipped in the fourth quarter of 2006. most people would think one hundred million requests would be for software patches to fix defects.1 percent) of the requests actually resulted in any new software fixes. Oracle’s Support group received one hundred million requests for service. How deep is it? Consider that 1. Just to drive the point home in 2004. support.9 percent) could be answered by knowledge in a library. or somewhere on the order of 25-50 terabytes of information. existing patches that were contained in a patch distribution system or in people’s heads—in short. nine-nine point nine percent (99. Yahoo and others have indexed) is thought to contain ten billion pages.030 petabytes (a petabyte is 1. Now.PLATFORM for OPERATIONS around defects—it’s around delivering information. at a minimum. and manufacturing information are all locked up in 151 .

and Autonomy. SAP.000 sites around the world. which was acquired by Microsoft for over $1B. which today probably exceeds $10B annually. have both focused on the challenges of enterprise search. the challenge is finding the relevant information. This is small when compared to the SQL database business. enterprise search in 2009 is today probably only a $500M business. Google launched their Enterprise Appliance in 2002 and. But relational database technology has no intrinsic business value (it’s just cool technology) and it wasn’t until the advent of enterprise SQL-based applications like Remedy. So why the gap? Perhaps the answer lies in tracing the history of SQL. and file systems all around the world. and many more to make the 152 . Today they are deployed in over 9. While seeing or indexing the information is one thing. While these efforts have all been successful. Peoplesoft. the second highest valued European software company after SAP. document management systems. Siebel. Independent companies like Fast. The opportunity to mine this deep Web has not been lost on numerous companies. Oracle was the most successful at commercializing SQL and by 1987 was generating $100M a year. has produced a major new release. every year since.CLOUD relational databases.

throwing them on the floor and using a search bar to try and read them.” and has been seen by many as Google’s attempt to compete with Wikipedia. So if the pattern is to repeat itself. and trying to read a book through a search bar makes little sense. we need to understand what it means to have search-based applications. Search can be so much more if you think of it as a technology platform. said “Google has been good and bad for search. and was lured away a few years ago to be VP of Engineering at Google.com. 153 .PLATFORM for OPERATIONS technology useful to business. he’s basically right. Good because they made search important. John Lervick. The term knol means a “unit of knowledge. who has been Chief Scientist at Yahoo. Chief Algorithms Officer at Amazon. once described the Internet in a very interesting way.” according to Manber. bad because they’ve made everyone think it’s a search bar. If you think about it. He recently unveiled a project called Knol. ripping all of the pages out. He said searching the Internet is like taking all the books out of the Library of Congress.” So what might these search-based applications look like? Udi Manber. Knol pages are “meant to be the first thing someone who searches for this topic for the first time will want to read. who founded FAST.

While 99.CLOUD While Knol is targeted at the consumer Internet. 154 . First. the application is read and write. Michael Rocha. It will take a new generation of these kinds of search-based applications to truly use the information buried in the Deep Web. Second. In that capacity. and their environments. he saw how the customers of complex applications struggled with being able to understand their software. has engineered a search-based application cloud service. as opposed to traditional SQL-based applications.9 percent of all of their service requests were met with known information. the search-based application acknowledges people add value to the information. a new company. their business processes. Openwater Networks. Finally. which only control which people have access to which information. While this has been a characteristic of all SQL-based apps few people think of search-based applications as having this capability. President and CEO of Openwater Networks. a team. was previously the EVP for Oracle’s $5B+ product support and services business. or a business unit. it was the cost and difficulty of delivering that information which inspired him to bring modern Internet technology to the problem. it standardizes the language (much as SQL-based applications have standardized processes) of a group. The search-based application fundamentally does three things.

automatable operations is the key to lowcost and high-quality customer experiences. How will you market the application. problems. managing operations is the most foreign to a traditional software company.PLATFORM for OPERATIONS But with greater and greater access to search technology. availability. sell the service. Summary Of all of the major functional areas. finance and organize yourself for success? 155 . We’ve now walked thru each layer of the cloud stack and now will focus the remaining chapters on the important areas for anyone building next generation application cloud services. there is bound to be hundreds of search-based applications just as we have seen hundreds of SQL-based apps. security and performance. But having a system of repetitive. whether in open source or in the cloud. This chapter has highlighted some of the areas in the management of change.

How is anyone going to know? Once upon a time you’d hire an “ad man” who would come up with an ad campaign and then find out if you could afford to run it in the Wall Street Journal. or just the Palo Alto Weekly. As a result many people think marketing means coming up with a clever tagline and finding a way to make it into a Super Bowl ad.MARKETING So let’s assume you’ve built the world’s coolest application and it’s going to revolutionize poultry management. or just the next major release of your ERP application. You’d have to make sure your tagline was short and sweet since you had no idea who would read it. This 156 .

This approach resulted in both founders maintaining significant ownership of the company. in 1996 and 1997. Min and Subrah financed their operations by having a group of low cost engineers implement software and plowing the profits back into funding WebEx (at the time called Active Touch). We’ll discuss the fundamental economics. In this chapter we’ll focus on marketing your product. but with the advent of the Internet and rich media. Economics of Sales & Marketing We’ll begin by considering the Webex case study by becoming students of their P&L shown in Figure 7.MARKETING might have made sense when the production and distribution of information was expensive. The funding line shows each round of funding they raised. we clearly no longer need to think this way. Consider the implication of this. which totaled $107M prior to their IPO in 2002. You’ll notice the explosive growth from 2000 to 2001. the single largest line item. You can also see that.1. given how we all view Web 157 . the traditional approaches and some of the newer ways for you to educate the buyers. You should note that Webex’s spend in sales and marketing is hovering around 35 percent. In these early years. This was the by-product of both a significant investment in sales and marketing and the explosive growth of Web conferencing post September 11. they operated at or near profitability.

Traditional marketing and sales has broken the pipeline into a series of stages.2 Traditional Sales & Marketing Pipeline 158 .1 WebEx P&L Traditional Sales & Marketing For those of you who have been doing high tech sales and marketing this section will be old hat. Figure 7. Figure 7.CLOUD conferencing as simple to use and simple to explain and simple to buy.

the big difference between consumer and enterprise sales and marketing is that there is not one decision maker – enterprise sales and marketing of imprecise. and web conferences is time consuming and further elongates the sales process. Ultimately. lunches. then leads. more expensive things is a team sport. Consider Figure 7. Different constituencies within the buying enterprise have questions that can only be answered by members of the selling enterprise (or customers that have already purchased – the reference customer). 159 .3 and talk to anybody selling business software. Scheduling the phone calls. Leads are both nurtured and qualified ultimately transitioning from a centralized marketing organization to the field where the potential customer is engaged. Even with a highly qualified lead the sales person finds himself or herself in the roll of traffic cop (or switchboard operator).MARKETING Early stages of the pipeline focus on getting awareness. No news here for any of you who’ve sold enterprise technology.

This is largely because the economics are so compelling. 160 . has been an area of considerable growth. Responsys have specialized in this area. Emailing to a list of 1. repetitive and specialized.CLOUD Figure 7.000 people costs $5 versus over $500 to send out direct mail. Everything from the number of characters in the subject line (keep to less than 50) to the time the email is sent out can determine the effectiveness. as an alternative to direct mail. As a specialist they focus on the details. We’ll focus on a few areas where marketing services have developed to improve sales productivity. Companies like Constant Contact.3 Sales is a Team Sport Hopefully thru CRM applications this process can be made more efficient. Lyris. Email Marketing Email marketing.

32% more applications were completed by the 161 . As a simple case study. triggered 30 minutes after the application was abandoned. and encouraged applicants to use phone support. provided a quick link to the partially completed application. It was executed in two phases. reminded applicants to complete the process. implemented an email program designed to remind and encourage prospective borrowers to complete their mortgage applications. sent one week later.4 Call at the right time e-Loan Story Scott Hilson. The program involves integration with the E-LOAN Web site to identify prospects that started but failed to complete the mortgage application. Siara Nazir. The first phase. Director. called the applicant’s attention to the company’s value proposition and also offered a quick link to the incomplete application. has helped many companies implement successful email marketing campaigns. As a result of this campaign.MARKETING Figure 7. The second phase. Built into the program was the ability to perform A/B testing on the messages sent to the customers. Customer Services at Responsys. Director of Online/Offline marketing at E-LOAN.

has some fundamental advice for anyone trying to optimize natural search results. This way. Blogs are great for generating traffic. we all know we click on natural search results more often than the ads on the side. saw a lot of traffic coming to their site because of their extensive network management content. you become the expert on the technology and the problem your company is trying to solve. who after years of experience in marketing Model Six companies formed her own consultancy. By paying attention to how Web pages are built and the content put in them. While there is plenty of advice on how to spend money on ad words. a network management company. Katy Roth. Solarwinds. you can create natural search results that drive traffic to your site without purchasing advertising (and more importantly be significantly more effective). • Build a FAQ into your Web site to capitalize on potential customers searching for reference or problem solving information. For example. Search Marketing A company’s Web site has become the face of the company so improving the volume and quality of traffic to a Web site from search engines via search results is clearly important. giving it higher weight in the results.CLOUD control group that abandoned but did get the automated follow up email. The content is generally considered by search engines to be fresher and more authentic. 162 • .

• Social Media Marketing Facebook’s early success with college aged people has caused some people to see social networks as just a tool to hook up or find old friends.300% this year. Today more than 15 million users create 1.500 “Tweets” every second. See the previous section on blogging. Yahoo and MySpace in total time spent online by Americans (6% of total time). a provider of monitoring and development products for Ruby on Rails.MARKETING FiveRuns. There is nothing worse than seeing a page with terrific content with a generic browser title. Facebook leads Google.com is a good example of a company that reworked their forms to pull in more community generated content. Since Februrary 2009. The number of users has grown from 60M users in 2008 to over 300M with the fastest growing demographic being people 35-55. They worked with their developers to tune their page generation engine to produce better metainformation that allowed their pages to come up to the top of natural search results. But they didn’t stop there. developed content on their blog that provided answers and guidance to Rails developers. • Have your community create content. But Facebook has become much more. 163 . iTaggit. Twitter has also enjoyed phenomenal growth growing 1. Work with a search engine savvy developer to make sure the meta content of your Web pages is optimized. Just check out Wikipedia for some good examples.

Within 30 minutes of the game starting the video servers crashed. Clara Shih attended the first Facebook developer conference in 2007. 164 . there were suddenly 100s of thousands of viewers. Now instead of anonymous cold calling. sales reps could get to know the person behind the name and title.com. Think of doing the same thing with a telephone and you’ll begin to understand the potential of these social networks. The application of these social networks to sales and marketing is still in its infancy.CLOUD Perhaps the best way to think about Facebook and Twitter is as a next generation telephone – and a very powerful telephone. and even ask for warm introductions from mutual friends. she developed Faceconnector which pulls Facebook profile and friend information into salesforce. As an example consider the recent case where Facebook simultaneously broadcast a soccer match in Spain for the Spanish Facebook community. With her friend Todd Perry’s help. Why? Because once one person commented and his social graph saw the comment (and started watching the game) and another friend commented and another.

product management. You have internal experts on the product in customer support. it was often cost-prohibitive to do this in any systematic way. Think about using social networks to market yourself as an individual. Claire Shih 165 . deciding the colors.MARKETING Clara’s counsel28 for those thinking about using social networking in your marketing efforts: • Use hyper targeting as an opportunity to test new audiences and see if your product can gain traction. Prior to hyper targeting. R&D. and consulting. • • Get Notorious This brings us to getting notorious. sales. Your profile information. quantity. Not only are your people sources of information. and friends all paint a picture of who you are and help to establish your brand. Customers of your product (the valued reference customers) that you metered out onto a conference call with a carefully selected 28 The Facebook Era: Tap Online Social Networks to Build Better Products. and Sell More Stuff. but so too are your customers. Use the social graph and knowledge about relationships across your audience to create a more personalized online community. and quality of the print that ultimately made it into a folder at an event at the Hyatt in San Francisco. photos. field service. But there are many more sources of information. spent weeks writing up a data sheet. Once upon a time the product marketing person interviewed the developers of the product. Reach More People.

No talking heads-show the product. By bringing more contributors into your marketing cloud. title. and phone number. should become notorious.” Reach out as an individual Name.CLOUD sales person and prospect today can have their testimonials videoed and made available on YouTube. A few bloggers including Vinnie Mirchandani. Make me think: “Wow. Encourage them to write. As you develop your marketing programs.Don't send a note from "your company. we will be able to find the knowledge we need delivered the way we want it—in a language and context that is accessible to each of us. While the old world of low bandwidth. that's impressive!” • 166 . in your third parties. Anshu Sharma. Blogging is personal. think about whom in your company. in your customers. Use video . Charlie Wood. or do Webinars or videos and communicate their expertise and more importantly their point of view. and Ed Herrmann provide some advice on getting notorious. or speak. expensive communication required that you needed to aggregate content in the Internet behind the veil of “the company” today we should be able to know what you think. • Be notorious . Limit it to 30 seconds. and what you can contribute.but not an overproduced one. what you believe.

5 Selling Spectrum As you begin to tackle this problem. Perhaps the greatest change is. Educate & Select Sales and marketing have been critical components of traditional software companies and new software companies. consider the fact that. photo essays. generally "this is so cool!"). very few people were sold a plasma TV. See what formats are popular. last Christmas. and use those. as they become more specialized. We’re 167 . (Top 10's. Figure 7. controversy. "best ever’s.MARKETING • Check out Digg . as we move closer to delivering these applications as a cloud service. and as the price point goes down. video. the traditional human network that has powered sales for the last twenty-five years.Look at the first few pages. we must re-invent the traditional account management model.

Our challenge is how we effectively sell imprecise items that cost $10. You only have to look at Amazon or eBay to see how successful this model has been. select the right answer for themselves. just that they were not bought because of a sales person.000.CLOUD not saying there weren’t a lot bought. Summary The Internet changes everything used to be a corporate tagline. We have also figured out how to sell low price (<$100). Consider Figure 7. they figured out the right size screen. The reason no sales person sold a TV is that most consumers of Plasma TVs did their research online. Perhaps this is what can happen in the future market for software. We should take a lesson from the Plasma TV. talk to experts and other users and. we have figured out how to sell expensive items ($1M) that are not very precise (SAP Manufacturing) and have grown large profitable businesses. in the end. by the time they were ready to purchase. Perhaps it is no more obvious than in the art and science 168 . Selling may be replaced by education and selection. As a technology industry. People will educate themselves on the good and bad of the solution. Purchasing became a transaction.5. had made their selection. read what others thought about the TV and. precise items.

169 . whether they find you through search. In the end. you still need marketing. With a low cost means of delivering notorious content. While communicating to people. the act of matching what a potential customer needs and what you have to offer remains the role of a quality marketing organization. in the way they want it. you now have an ability to deliver personalized information to every member of the team that influences a purchasing decision—when they want it. email. no matter how simple we try to make the software. or social networks.MARKETING of marketing.

After about a year. Finally. and he scheduled a meeting with the head 170 . Rudin. was asked by Tom Siebel to lead the efforts for Siebel On Demand. There was a sense of urgency. who today is the CEO of LucidEra. it doesn’t take too long until someone says.SALES While SaaS or cloud computing might be all the rage.com was winning more and more business. of course. wasn’t sure if it was a good idea. “But this will cannibalize our existing sales. Ken Rudin. Rudin was able to work out the product strategy and business model.” Sometimes it’s stated even more clearly. as Salesforce. Early in 2002. but Siebel finally convinced him. he was ready to tackle the sales and distribution channel.

that’s an outstanding plan. But. but stay the f—k away from my accounts!” This chapter is dedicated to sales and distribution. said. moving existing applications to delivery as a cloud service with an existing channel and product (Models Four and Model Five). the challenges will pertain to developing a new sales force in a world where the economics don’t bring you million-dollar deals that can fund the creation of a channel. In Models Four and Five. The VP listened politely to Rudin’s presentation and. there are many challenges. Does it mean the comp plan will change? Is this new on-demand group going to compete with my sales force? How important is lead generation? In this chapter. In Model Six.SALES of North American sales. you will need to figure out how to leverage your existing sales force. the solutions to the challenges here will make or break any new business. we’ll address some of these issues and also let you hear from some successful sales and marketing executives. Transforming Traditional Software Sales Every Model One software company worries about cannibalization when they consider a move to delivery of their 171 . More than with any other group. your primary advantage is that you already have a sales channel. For the traditional software company. “Ken. at the conclusion.

The only change is they have a new delivery option (one which they can be compensated for). hardware. Joe says. third-party software. organize the sales teams around current delivery capabilities. “Don’t sell futures!” Finally. the existing sales force continues to sell the software as it always has. who led sales for Oracle On Demand for seven years and today is Group Vice-President of JDA’s on demand business.CLOUD software as a cloud service. nothing changes. His first piece of advice is to make sure you find a team of sales professionals that have a mix of experience in selling both traditional software products and services. it’s 172 . Whether that’s a perpetual or term license. Ensure that the sales team is setting proper expectations with the customers and selling a solution that your company has the ability to deliver on today. etc. train the sales reps to find customer challenges that can be solved by having you provide the complete solution without letting the customer dissect the solution into its components.. up front or expense it over time? Joe King. Unlike selling traditional software. In other words. If you adopt Model Four. has first-hand experience in how to sell when you’re in Model Four. It doesn't have to be this way. Do you want to manage our software or do you want us to manage our software? Do you want to capitalize and pay for the software. Next.

000 (for the first year). the average selling price is going to be significantly under $100. And when you get to the CIO or IT director. In Model Six. There still needs to be value in your software in solving a particular business problem. the traditional sales person who manages an account and makes a good living sitting in a few accounts is unaffordable in the new world. they will ultimately be better and cheaper at insuring security. so no longer can you fund the software sales by the $1M up-front license fee. or identifying performance problems than you as a single isolated island having to figure it out for yourself. Selling New Application Cloud Services For those of you building a Model Six business selling software as a service is in many ways no different than selling traditional software in Model One. 173 .SALES important to discuss the underlying infrastructure that supports the solution. What is different is the economics of the sales process. As a result. managing an upgrade. who will inevitably say: “I can do this” (because they have). but remember the customer is not buying servers and data centers separately from your complete solution. you need to be armed with the evidence that when a software company delivers the service in a repetitive standardized way.

WebEx. A Model One software company offered to do a free six-week discovery process and scheduled a demo for eight weeks later. SpringCM. the good news is Model Six products are reasonably precise versus the large application footprint of traditional enterprise software. QuickArrow. Remember.” the idea of scheduling custom demos is a thing of the past. Stratus went with a Model Six solution and deployed the whole system in seven weeks— worldwide. and Postini. Now with adequate Web marketing and standard demo deployment. a $250M a year computer company. Joe relayed a story of how different traditional software companies are from new Model Six companies. Salesforce. there are some things that are better. The 174 . is a big user of application cloud services. Halogen.com. some group of customers can try your product and then proceed to use it in small work groups.CLOUD While that aspect is worse than our old world.com and WebEx both can attribute their high growth and early success to having products that could be used simply and economically by small groups. CIO at Stratus Computers. Today he uses Salesforce. Xactly. Ceridian. Since the software is either “always deployed” or “ready to be used. Eloqua.” or “quickly configured. Joe Graves.

SALES bad news is that the average selling price is considerably lower in Model Six than in Model Four. Every lead from this marketing is categorized into three buckets. Salesforce. Aaron Ross was Director of Corporate Sales with 175 . the next level within 72 hours.000 in marketing per sales rep per year. and email campaigns. With an eight-to-one conversion ratio and a fifteen-day sales cycle.000. Salesforce. with an average price per seat of around $70 per month.com has also had a few tricks up their sleeve. you need a sales machine. And when you need volume.000 per year with 50 percent of their customers having an average yearly contract of less than $1. daily demos (scheduled and executed from India). High priority leads are responded to within 24 hours.000 per year. So from a company perspective you need volume. WebEx’s approach has been to spend nearly $100. A sales person can no longer depend on landing a few big whales to make quota. Webinars. you can see they have perfected a sales engine. This includes free trials. The point of both of these numbers is that an average deal size is nowhere near the average selling price of traditional software. meaning that their average yearly contract is under $20. WebEx’s high-end customers average around $25.com averages about 20 seats per customer.

Vendors can’t control the flow of information. think about how much information and control over the 176 .” In the past. generating more leads. The best strategy is to embrace this.” This was possible. Today.com from 2001 to 2006. Today. which sales people close. hiring more sales people is not the main engine of growth. Likewise. There just are not enough great sales reps to hire out there. He started by saying something that might seem counter-intuitive: “The size of your sales team does not drive growth. prospects can find out as much as they want about you before they ever call you.” Historically. who would work their patches to generate leads and close more business. Ross went on to say: “Give up Control. and instead. instead of obsessing over the number of sales people in your team. the sales organization wanted “control over the sale. is what drives growth. because it was very challenging for customers to get unbiased information about vendors. obsess even more over your sales productivity. so helping your current team do more with less is as important as hiring. We asked him what advice he would have for establishing high velocity sales. a VP of Sales would drive growth by hiring more sales people. The Internet changes everything. In Ross’ opinion. that doesn't work.CLOUD Salesforce. Instead.

Most VPs of Sales focus too much on the second half of the sales process: sales deals in the pipeline and what can close.. 177 . So instead of “Always be closing. they’re terrible at it. and anything artificial irritates them.. Yet.” focus on building a sales team that is constantly asking. Finally. “Are we a match?” You can't afford to waste much time (money) on prospects that aren’t a great fit. they hate doing it. the more investment sales put into both the quantity and quality of leads.SALES sale you can put into the hands of the prospect. They tend to ignore the first half of the process: lead generation and lead qualification. Here are two specific examples of common mistakes: not hiring enough junior sales people to qualify inbound leads.making sales results easier. How much can you educate the decision makers and influencers? Ross’ next piece of advice again might sound counterintuitive to old-school sales people: “Stop the ‘Always be closing’ mindset. lead generation and lead qualification are as important as closing. the higher the quality and the larger the sales pipeline will be. and as soon as they get some pipeline. One way to fail at generating leads is to expect your own quota-carrying sales reps to do it. they become too busy to prospect anyway. and not creating a dedicated team that does nothing but outbound prospecting.” Buyers have seen all the sales tricks and programs.

In Q1 2007. Forrester reported companies based in Europe. Since the earliest forms of selling software as a service were large outsourcing deals (Model Three). When asked about the duration of renegotiation. there is a cottage industry in providing advice to companies negotiating contracts. the Middle East.7B. and Africa (EMEA) signed 75 large outsourcing deals in the quarter—with a total deal value of almost €5. They published a paper providing advice to customers renegotiating outsourcing contracts. many contract issues for Models Four through Six carry with them some of this legacy. you might discover the same thing that Bechtel’s CIO. 85 percent of those surveyed indicated the restructuring process 178 . TPI is a leading global outsourcing advisory firm. The outsourcing (Model Three) business is still a growing business around the world. Hank Leingang. If you were to look at the contracts for those deals.” With this degree of complexity. found when he plowed through its contract to outsource desktop support and network management to Electronic Data Systems: “It was three inches thick.CLOUD Contracts Contracts vary significantly from model to model.

contracts like these SLAs are heavily negotiated. other vendors require a two-year commitment. Companies like Salesforce. If you want to spend a couple of frustrating days. you can lock five experts in a room and try to define what uptime means and what penalties you should pay if the uptime is not met. Typical Model Six terms allow customers to pay on a month-to-month basis. Service Level Agreements In almost all of the contracts from Models Three through Six. so contracts can be terminated with 30-days notice. or a Tool to Increase Value? TPI Report. take a look at the Yahoo terms of service or the WebEx contracts. some vendors choose to enforce a penalty against the total value of the contract if a customer wants to cancel the deal prematurely. 179 . January 2007. Fifty-five percent finished renegotiations in less than six months. In Model Three. and 13 percent finished in less than three months.com offer SLAs on contractual terms on a case-bycase basis. such as Blackbaud On Demand. the subject of SLAs (Service Level Agreements) will come up. Furthermore. Binding SLAs to contracts can vary. In other cases.SALES required less than a year to complete. For examples.29 At the other end of the contract spectrum is the online click wrap agreements. the 29 Restructuring Outsourcing Agreements: An Indication of Failure.

Although the financial penalty might seem minimal to the individual customer. As noted earlier. In either case. so you should first focus on what service level your operations and development group are engineering to deliver. So the total financial incentive for Taleo to keep the system up and running all day and every day is enormous. change. they point out.CLOUD contracts stipulate 99. if they are writing a check to that customer. you can decide to what extent these metrics are embedded in a contract. “We occasionally modify the terms slightly for customer’s specific needs. simply called the “Nordstrom’s guarantee. chances are they are writing a lot of other checks.” Service levels in availability. problem.” Taleo provides penalties in their contract if they don’t meet certain uptime standards. but it is a competitive agreement. After that.” said Blackbaud’s CEO Marc Chardon. Oracle pioneered a different approach. and performance management were 180 . security.9 percent uptime and are a standard part of the contract. There is a 99. the important point is that a user of an application wants the service to perform. as this agreement governs what we do operationally. “We're very strict on ourselves.9 percent uptime guarantee and a commitment on the specifics for data handling and server maintenance.

Security Management – Have you completed a SAS 70 audit? How frequently are backups taken? Where are backups stored? How fast can data be restored? Under normal circumstances or in the case where there is complete site failure because a disaster has occurred. However. and 99. Let’s highlight the major areas you should focus on. and do that for planned and unplanned outages. 181 • • . You should consider defining availability by total number of outage minutes per month. The Nordstrom’s guarantee was an effort to separate the issue of penalties from a description of the service level metrics and objectives.5 percent.99 percent… They all sound about the same.SALES published and reported. per year. no questions asked. or will you stage them over a three-month period? Availability Management – Many contracts call out percentages of availability (99. per quarter.9 percent. don’t they?). It will cause you to focus on how rapidly you recover. less than a handful of rebates were paid out. if the customer was not satisfied with the service for any reason. Writing these down and measuring them is key to improving the overall quality of the service. In four years of offering the Nordstrom’s guarantee.7 percent. • Change Management – Define how often upgrades to new releases will occur. which is a much more important metric. they could request a rebate of 20 percent each month. 99. 99. Will you move all customers at once. This begs the question of whether this refers to unplanned or planned outages.

• You would be wise to separate the definitions of your service levels from the legal and contractual side. be proactive about the service levels you will provide and publish them to your customers. Become a student of WebEx’s and Salesforce. but providing some reference implementation to benchmark on a release-to-release basis will help you provide performant apps. just look at the upper right-hand side of your Google search results when you search for “SaaS”. and if the solutions 182 . “If you create a level playing field in the sales compensation. For a service level that is reported on every transaction. what level of service will you be providing? Performance Management – What response time should customers expect? How much disk space is included? How quickly should reports run? This will be a difficult area. “We try as much as possible to make the sales rep indifferent to selling an on-demand or on-premise version. Some will push you towards EDS or IBM Global Service agreements. In the end.CLOUD • Customer service management – When a customer has a question about your software (these could be defects or just usage). remember the customer wants the service to work. Instead. points out the universal truth in effective sales compensation: You’re looking for sales people to respond to the metrics that you give them. Sales Compensation Marc Chardon.com’s contracts and terms. CEO of Blackbaud.” he said.

and consistently with the plan type and organizational style. transparently. Credit directly for sales compensation. One good (transparent) report delivered to payees at the right frequency keeps the focus and motivation where it should be. • • • • • 183 .” When you look for expertise in sales compensation you’d have to consider the team at Callidus Software. Create a transparent dispute resolution process. Report regularly. Analyze and change on a predictable schedule. sales reps should be. Even if you are able to follow every piece of advice above. and usually are. • Keep your compensation plans straightforward. The smartest and most successful companies are constantly evaluating their sales performance data and are not afraid to make changes to compensation plans. Make sure you have the data—and that it’s clean. Many companies come up with plans with dozens of variables and multiple levels of intricate dependencies. They have some guidelines that are applicable across a number of our seven business models. errors and problems will happen.SALES are roughly as easy to sell. good at evaluating the shortest time necessary to get their commission. The most effective plans credit the primary seller along with those directly connected to that seller in the chain of command.

While traditional sales has been on trying to know something about a customer you don’t have – for the first time companies can now focus on learning more about the customer they do have and based on that information personalize the offering using that knowledge. each taking the same amount of time (1 tick of the clock) then instead of 35 ticks to get 7 instructions executed you could do it in 11 ticks. Nowhere is this more obvious than in the case of Amazon. 5 parts. Years ago every instruction took 5 clock cycles to execute. Computer architects have struggled with how to make computers faster and faster.CLOUD Installed Base Sales Most of the focus on selling has been on getting the new customer. But over 10 years ago some smart guys figured out if you could separate instruction execution into 5 stages. less precise products? Here we can take a lesson out of the notebook of computer architecture. again a by-product of selling a product for large upfront money versus driving revenue on a monthly or yearly basis. While turning up the clock speed is one answer. another answer lies in being more efficient about how an instruction is executed. So with 184 . as a result seven instructions would take 35 ticks of the clock. Below is a diagram showing the typical execution of an instruction. But how can this be applied to higher ASP.

. Figure 8.SALES the same clock speed the computer was in the limit five times faster.1 Pipelining So what does this have to do with sales? Our marketing & sales pipeline is no different only the clock ticks are not so fast. one instruction. 2 weeks) and then move customers from one stage of the pipeline to the next.g. Then each stage could be optimized and specialized and over time sales transactions could be much more parallel than they are today. But indeed the same principles can be applied. The trick is to engineer each stage of the pipeline to last a fixed period of time (e. before we start the next.000 companies who already owned Oracle applications. The conventional way of managing the pipeline is no different than the traditional way of instruction execution – we finish one deal. and were applied to sell a standardized Model Four service to 10. Anyone selling to a large installed base so reconsider how – and not just treat it the same as potential 185 .

Unfortunately.CLOUD customers – where you have no idea who they area.000. Names like Ingram Micro. Tech Data and CDW. Indirect Channels The last major generation of computing was called clientserver computing. you are now in competition with these traditional channel players. they can afford to pay the sales 186 .000. today all multi-billion dollar companies. While the reseller saw the value of a Model Six business. Compaq. In early 2006. Its rise was not only driven by the availability of low cost hardware (Dell. With $70. I had the opportunity to spend a day with one of Microsoft’s largest resellers for its Great Plains Software.000 license. etc. nor what experience they’ve had with the product.) but also by the creation of a new channel of distribution. etc.) and low cost software (Microsoft. This client-server indirect channel has been built on delivering “boxes and bodies” to implement and deliver the application. Model Five or Model Six. they were as addicted to the “perpetual license drug” as the vendor. Allow me to explain. or $70. they get 35 percent of the sales price. whether you’re in Model Four. Corel. Today when they sell a $200. provided a low cost indirect channel that matched the price of the technology products.

have some potentially big challenges ahead. Based on interviews with SaaS providers. Darren. IDC believes those new solutions providers will emerge that are more business process oriented rather than • • 30 TenWolde.SALES rep to drive the sales cycle and then perhaps make more money by doing the implementation of the software to the specific and unique requirements of the customer. Thirty-five percent of that number is now only $17. September 2007. That’s $50.500—and the implementations are far more standard. Key highlights from their study were: • There are still nascent partnering business models in the SaaS ecosystem. This will have implications on the valuations of VAR businesses. Erin TenWolde and Darren Bibby of IDC published a research paper 30 in late 2007. Indirect channels have been important to the software business. Erin and Bibby. which traditionally have been valued based on a multiple of revenue. 187 .000 might be spread over four years.000 per year. IDC Report #208547. With many traditional IT challenges removed with SaaS. there is still a lot to learn and most of the activity is occurring on a trial-and-error basis. especially in terms of securing top-line revenue in resale scenarios. Traditional channel partners. Now the $200. particularly value-added resellers (VARs). so it’s no wonder people are trying to figure out what a next-generation channel will look like. Consider the same scenario in subscription view. The Emerging SaaS Channel.

who founded Computer 2000. Ashton-Tate. Compaq was founded. They started selling in companies who never even wanted a large 188 . Those partners that can develop discrete expertise in niche areas will be well positioned to take advantage of the vast SaaS opportunity. and Intel? An interesting. This was a tough strategy and a cumbersome fight against established computer manufacturers. He saw the first three to five years of the PC industry focused on replacing existing midrange computers like IBM /34 or DEC PDP 11 with personal computers. most importantly. He’s uniquely positioned to discuss what a next-generation channel might look like. Schultze remembers 1981. Axel Schultze. almost unnoticed change happened: young startups entered the market and didn’t sell into the established businesses. the largest reseller of PCs in Europe. and Microsoft introduced BASIC. Can we learn from the PC industry development in the 80’s? What happened back then that led to the explosion of the PC industry and the enormous rise of companies like Compaq. when IBM announced the PC. established consulting processes and. established sales channels. IBM. also founded a Model Six company called Blueroads. Microsoft. Lotus.CLOUD technology focused. but into a new market segment—companies below $50M in revenue. established software.

a channel emerged that helped sell and install those PCs and PC software for $1000—a price a mid-size systems integrator wouldn’t be bothered with. What is missing are the new channel companies. Axel boiled it down to a simple premise: “There is interesting software for a large variety of customers and industries.” Will history repeat itself? 189 .SALES midrange or mainframe computer. And along those lines.

the Magic Number and how much it takes to launch a new business. churn. but if you don’t have money (and people) it will be hard to turn it into a reality. We’ll discuss monthly recurring revenue (MRR). revenue recognition. 190 .FINANCE You might have a killer application and a market that is rapidly growing. This chapter focuses on the money – some of the key metrics that you’ll need to track in order to insure a long running profitable business.

the largest part of revenues stems from vendors’ license fees associated with software. but logged in accounting books differently under Model Six. If a Model One company sells software for $1M. In the traditional software business. assumes that they choose to move to Model Five with their existing license business versus working in Model Four. traditional (Model One) software companies have been concerned about experiencing a negative short-term impact on license revenue. The same dollar amount is spent on applications.000 per month for the next 10 years. Traditional software revenue accounting requires that the vendor’s fee be allocated to the various elements based on something called vendor-specific objective evidence (VSOE). This requires the deferral of 191 . This. VSOE is limited to the price charged by the vendor for each element when it is sold separately. they may get $20. they recognize the million dollars on the day they sell it and it goes into the revenue for that quarter. A Model One company recognizes revenue from license fees when the software is shipped to the customer. of course. As a result.FINANCE Revenue Recognition Many traditional software companies are concerned about moving to Models Four through Six because of something called ‘revenue recognition’. If they were to sell that same system as software as a service.

Contracts are yearly. the entire fee can be recognized ratably over the term of the customer support contract. subject to renewal by your support renewals teams. one-time charges. such as implementations or customizations. In addition. can be treated the same way you have managed professional services contracts. former head of finance and business operations for Oracle On Demand says. some of the same contractual terms and sales methodologies can be used. “For those pursuing Model Four. this should be no different than how this is done in your traditional software business. but. recognition of revenue must be deferred if undelivered elements are essential to the functionality of any delivered elements. you’ll have to recognize revenue based on milestones and the level of work that has been completed. Priscilla Morgan. In these cases.CLOUD revenue until VSOE can be established for all elements in the arrangement or until all elements have been delivered. You’ll also find out that your services will be much more specific than the terms you already have for product support. the simplest way to think about providing software on demand is to treat the revenue the same as you have treated customer support. once again. If customer support is the only undelivered element in the arrangement. however.” In fact. This way. 192 .

9 million ($50k * 78) of revenue over that year. MRR is a powerful idea but as such has both upside and downside. then the business is close to covering fixed operating expenses under the $50k MRR bookings scenario. [Note: 78 is simply the total of the revenue months during the year. CEO of Zuora.2M in year 2. since in the last month the company was doing $600K in MRR. which by the way. should also be used by Model One companies to track their support revenues. Furthermore – next year if it sells no new business revenue the company will still grow to $7. A measure used by many Model Six companies is MRR (Monthly Recurring Revenue).FINANCE Monthly Recurring Revenue (MRR) Tien Tzuo. And the company just missed its revenue number by $2 million. if your fixed operating costs are $600k per month.95 million. but now 12 months away under the $25k MRR bookings scenario. Worse yet.. On the upside assume your company gets $50k of new MRR during your first twelve months of revenue generation. assuming no 193 .. But what happens if the company does only $25k of MRR bookings.+1]. points out that you’re going to need the right metrics to run any Model Six business. Now the first year revenue is only $1. 12+11+10+. The rule of 78 says the company will generate a total of $3.

Traditional Model One businesses have been great since with typical software license being sold for an Average Selling Price of say $500. Unfortunately. Which of course brings us to the subject of cash. which historically have 194 . However.CLOUD churn. Cash As Subrah Iyer commented. Figure 9.1 Power of Payment Terms Cost of Sales: When to Hire for More Sales As we’ve repeated Model Six companies don’t have the advantage of Model One companies.000 could bring you the cash you needed to hire more people or buy more computers. this is not exactly the case in a lower ASP. Model Six business model. Meaning you’ll need another $2M on cash/investment. if you could get a customer to change their payment terms and pay for the service a year in advance or better yet three years in advance it would do a lot to help your cash needs.

FINANCE derived a large up front.000-server cloud for its 5. To get a sense of the cost of service. Moreover. which force the cost to be expensed upfront. Omniture operates a 15.2 that produces what they call a Magic Number. In 2005 Omniture’s negative free cash flow was in excess of $22M—more than its total annual revenue—and it had spiked higher at some points. But when is it the time to step on the gas? When is it the time to ramp your sales force? Josh and Omniture have been relying on a simple calculation. 195 .” he said.” said Josh. “When we were really stepping on the gas in 2004 we were GAAP unprofitable. shown is Figure 9. The financial consequences look exceptionally dire when expressed according to GAAP accounting rules. Even when management understands the game plan and has the full support of its financial backers.000 customers. it costs us more money that quarter—it costs us more cash that quarter” explained Josh James. and processes almost a trillion transactions per quarter. license fee when a sale is completed. it takes courage to plow ahead regardless. the service requires significant infrastructure investment. “Every time we add an incremental customer. “It is a scary moment. CEO of Omniture.

for any given period. or its opposite. renewal.3 Magic Numbers for Model Six Companies Churn: Cost of Keeping a Customer Another critical metric for recurring businesses is churn.CLOUD Figure 9.5. invest in more sales people and if the Magic Number is less than . what percent of the business 196 . But there are many ways to calculate churn. For example. According to the analysis any time the Magic Number is . there’s probably something wrong with your business.75 or more. Figure 9.2 Magic Number Calculations Figure 9.3 shows the analysis for a group of companies.

breathing representation of the relationship between you and your clients. Perhaps you started off with 10 licenses of an application. In a Model One software company-you express interest in purchasing my products. I can measure how much of my service you are using and compute your bill. what percent of the business that is up for renewal actually renews? Financial Systems Many companies dramatically underestimate what it takes to run a Model Six business. and the transaction is completed.FINANCE at the start of the period is still there at the end of the period? The part that is lost is considered the churn. ship you the products. In addition. Each month. send you an invoice. and you want to suspend delivery for 2 weeks while you are on vacation. the customer may choose to change the service. Not so in the recurring revenue world. and now you want to add text messaging. at any time. and every change to the subscription must be handled by all your 197 . For a given period. The first order is only the start of the relationship. and now want to add another 2 licenses. a renewal metric may be more appropriate. Perhaps you have purchased a cell phone plan. Subscriptions are a living. I write up an order form. Or perhaps you subscribe to delivery of the New York Times. collect payment. For companies with longer-term contracts. such as a month.

learned early on the importance of having the right pricing & packaging strategy in order to grow a subscription business. having the right billing and payment systems is important to growing and scaling the business. and we thought that would be the price forever. We priced our sales force automation (SFA) service at $50 per user per month. today CEO of Zuora. to billing. we thought the right strategy was to keep our pricing simple. At Salesforce.but in return. they wanted more features. whether you buy or build must give you pricing and packaging flexibility. though. For any Model Six company. Different companies had different needs. The market. and through to collections. but they told us that $50 a person was too steep a price. “When we first started. That's when we embarked on our 198 .CLOUD back end systems from quoting. to order provisioning. and different price points. Many tiny companies loved the idea of an on-demand service to manage their sales force. Tien Tzuo.com. and stronger customer service. Any billing system. At the other end of the spectrum. had other ideas. large companies like Autodesk told us they actually wanted to pay more -. We quickly realized that a one-size-fits-all pricing model would never work.

Customers want choice as to how they consume your services -. and introduced 4 new price plans called "Extra Value Plans" for its heavy users.99/month. Today. Zipcar started off with a $50/year membership fee for their popular car sharing service. you see the same evolution in their pricing models. Netflix started off in 2000 with a simple model -. an Enterprise Edition at $125 per person per month.$19.FINANCE packaging strategy that ultimately led to a Professional Edition at $65 per person per month. and a Group Edition at $995 per year for 5 users.and how they pay – and the financial systems for service businesses need to reflect this degree of flexibility.99/month Similar.95/month. The basic lesson to learn here is that different customers have different needs. Their most popular plan is $16. Fast-forward almost a decade. 199 . Zipcar also has renamed the original plan as the "Occasional Driving Plan". But subscribers new to Netflix can dip their toe in the water with the $4.” If you look at other subscription companies like Netflix and Zipcar.99/month 1-DVD-at-a-time plan. and Netflix now has over 9 plans hitting multiple price points. and heavy users can upgrade all the way to the 8-DVD-at-a-time plan for a whopping $47. labeled as the 3-DVD-at-a-time plan.

The first thing she looks at is total available market. a William Blair & Co. her third criterion is to evaluate the level of competition: the more competitive the market.” And. it’s “one of the reasons Salesforce. and Concur have all done so much better than other players in the market. Vocus. So what are they going to be interested in? How do they evaluate investments for the public markets? Laura Lederman. the financial analyst community will be talking about you. From her perspective.” Number two on her list is the quality of management team. Venture Capital If you were a student of Salesforce. Of course the money was well spent since particularly in the later stage 200 . The larger the market.com. you’re going to hope that. the better the investment is likely to be. This is also the same type of number in Webex’s case. financial analyst who focuses on new software companies. typically the more difficult the investment. one day.CLOUD Financial Analysts As a new company. sees three important evaluation criteria. “It’s the most important factor. finally.com and you’d know that it took $60M+ in capital to get the company to an IPO.

com have high valuations relative to sales and earnings. an accounting software company. Winblad discussed her 201 .FINANCE investments there was a direct correlation between investment in sales and revenue. In 1998. Again.. like WebEx. Ann Winblad began her career as a systems programmer. Inc. she co-founded Hummer Winblad as the first venture capital firm to invest exclusively in software.4 Salesforce. and in 1976. she co-founded Open Systems. Their first investment in a Model Six company was Employease in 1998. In a recent lecture at Stanford. and on a much slower growth curve. with a $500 investment. CRM took about six years to reach $100M in revenue.com P&L Some of you reading this are starting up new companies and contemplating being “like Mark”. Students of traditional software companies know most of them struggle to get to $50M. Figure 9. It is no wonder that both WebEx and Salesforce.

being able to address unserved markets. and R&D can come together under a unified business model. operations. 202 . and always “serving” the customer are all characteristics Winblad sees as being better. being independent of capital or IT budgets. multi-tenant architecture Can be sold to small and medium size businesses (SMBs) as well as enterprises Requires little customization Requires low professional services Pricing model is subscription based (by time period. and identified a number of advantages.CLOUD view of software as a service. Having forward predictability. the growth rates and valuations can be astonishing. But if the sales. Her key characteristics of a Model Six company are: • • • • • • Can be sold and delivered over the Web (via browser) Single instance. user or transactions) Summary Finance and business operations for a software product being delivered as an on-demand service are very different than what we’re used to in the traditional software model.

Whether it’s cloud computing. If you’re an existing software company. and insuring you have the right people in the right jobs. Web services. If you’re a new software company. In this chapter.HUMAN RESOURCES Software is a people business. The challenge of growing a new business is difficult and getting the right team and the right model will be key. you’ll face the Innovator’s Dilemma. we’ll discuss some of the people challenges: finding or teaching the right skills. establishing the right organizations. finding the right people and organizing them 203 . software as a service. or software on demand it makes no difference.

1 how the company’s growth was tied to it’s ability not only to raise money. Geoffrey Moore’s landmark book. Right People.CLOUD appropriately will be a constant challenge. best describes the lifecycle of a high tech company.) and performance management (Successfactors. In this and in subsequent books. and lots of people. all of this begins with people. etc. Figure 10.1 Hiring ramp at Webex And by the way. but also to fire people. Kenexa. as you can see in the year 2001. Crossing the Chasm. Right Stage For many of us. Economics If we again use Webex as a case study of a successful company you can see in Figure 10. But. etc. but also to hire people. As a result much of the HR application cloud services have focused around recruiting (Taleo. not only to hire people. Moore describes the seven major stages of a successful venture: 204 .).

learn and change as they cross through each of these stages.HUMAN RESOURCES • Early Market – A time of great excitement when customers are technology enthusiasts and visionaries looking to be first to get on board with the new paradigm Chasm – A time of great despair. We often see that the CEO of a company at the early market stage is not the CEO of the company when it reaches Main Street. when the earlymarket’s interest wanes but the mainstream market is still not comfortable with the immaturity of the solutions available Bowling Alley – a period of niche-based adoption in advance of the general marketplace. it’s important to realize that people need to go through these stages as well. driven by compelling customer needs and the willingness of vendors to craft niche-specific whole products Tornado – a period of mass-market adoption. 205 . when the general marketplace switches over to the new infrastructure paradigm Main Street – a period of after-market development. there are the rare exceptions of the individuals with the gift of being able to adapt. Names like Jobs. and Gates are the clear exceptions to the rule. when the base infrastructure has been deployed and the goal becomes to flesh out its potential Assimilation – the technology loses its discrete identity. moves into decline. Of course. Ellison. and is supplanted by a new technology paradigm • • • • • While successful companies need to go through these stages.

it might look like some companies took a straight line from inception to Fortune 500. The mistake companies often make is putting an operationally excellent leader in place before they’ve crossed the chasm. the title CEO does not mean Chief Executive Officer. Getting the right people at the right stage. that’s the ticket. what are the characteristics that are most important in each of the stages? Tom Kippola. instead it should mean Chief Experiment Officer. there were many left-turns. Of course. Making the quarterly sales numbers. co-author of The Gorilla Game. most would have failed early in their lifecycle. and sticking to a budget are all requirements for any successful business. and then press on. delivering releases on time.CLOUD But for the rest of us. 206 . there does come the day that experimentation and cheerleading need to be balanced with operational excellence. recently said. Without the ability to quickly run experiments.” While. right-turns and U-turns along the way. “As a company is trying to cross the chasm. learn from them. the reality is. or forgetting to get their operational house in order as they’re headed down to Main Street. in hindsight.

development people that are different from what we’ve seen in the traditional software world. operations people who understand both software development and software delivery. As Evan Goldberg pointed out. People who have had to optimize performance or availability of software implementations in the field will have been exposed to some of the key processes and challenges involved with delivering Web applications. or vice versa. but having the right people at all of the skill positions is the key to success. you’re going to need what he calls next generation DBAs.HUMAN RESOURCES Right People As with any team. Often times. it forces conflicts between the race for new features and the ability to manage change in a reliable way. Clearly. Lynn Reedy recommends you have the VP of R&D work for the VP of Operations. these individuals are leading premium support services and are located 207 . your engineering group needs to both have a “high tolerance for change” as well as be “accurate and careful programmers. So.” Furthermore. By doing so. where are you going to find these skilled people? One place to look is in the traditional software product support organizations. Delivering cloud services requires operational people. having a good coach and quarterback is essential. this is something we never had to do in traditional software companies.

Many of these people have run large-scale operations. the leading online shoe retailer he made culture a cornerstone of the company. the people who ran American Airlines SABRE). So it should be no surprise that when he invested in and later became CEO of Zappos.CLOUD at or near a customer site. was recruited out of Wal-Mart.g.. You may also discover some of the greatest expertise to be individuals who have been in end-user operations where they have had to manage applications for a large internal population (e. Rick Dalzell. Internet advertiser LinkExchange. 24-year-old Tony Hsieh sold his company. to Microsoft for $265M. who recently retired from Amazon and was their Senior Vice President of Operations and CIO. Culture starts from the hiring 208 . While LinkExchange had been a successful company. You’ll have to find the people that can generalize the lessons learned in the field and apply them in a standard. the common values shared by everyone in a group. he never focused on the company culture. uniform way. The challenge for these people is that their tendency will be to customize a solution for each customer installation. What some people don’t know is he sold it because he stopped wanting to come into work. Culture In 1998.

Be humble 209 . creative. they are offered $2. Create fun and a little weirdness 4. Pursue growth and learning 6. Embrace and drive change 3. and open-minded 5. Do more with less 9. It’s several hundred pages of employees from all across the company describing what the company culture means to them. is designed to ensure employees are there for the right reasons." Zappos. Build open and honest relationships with communication 7. where they ask. In 2008 less than 1% of the trainees accept what the company calls "the offer.com also publishes a "Culture Book" every year. After the training. “Which superhero would you want to be?” and continues thru their four weeks of training where every employee (even VPs) answers phones in the call center and ships shoes out of their distribution facility in Kentucky. which started at $100. Build a positive team and family spirit 8. The company itself maintains 10 core values: 1.000 to leave the company—no questions asked. Be passionate and determined 10. Deliver WOW through service 2. Be adventurous. This "quit now" bonus.HUMAN RESOURCES process.

"We want people who are passionate about what Zappos is about . including finding out a place to get food. Concur is a unique case study—a company that made the transition from a traditional license model to software as a service: from Model One to Model Six as a publicly traded company. Buy New means to acquire companies with Model Six business models and Remodel refers to adopting a Model Four or Model Five business in parallel with your traditional Model One business. "Our number one focus is our company culture. The Scrape strategy is best illustrated by Concur. when the hotel had closed room service.service. Buy New and Remodel. Scrape For those that are attempting to transform existing Model One software companies there are many unique challenges." This passion has translated to hundreds of stories of how Zappos employees have delivered unexpected service. Scrape refers to an “all-in” approach to moving from Model One to Model Six.CLOUD Tony says. You can see the 210 . Figure 10.com. We’ll call these Scrape. You have three fundamental organizational approaches." For Zappos. I don't care if they're passionate about shoes. a great culture translates into great service.2 shows you the transformation in revenue terms.

with the global meltdown of 2008. along with the growth of the subscription line. you’d see the price crashing below $1 a share. CEO Steve Singh says Concur was able to do this because the 2001 Internet Bubble burst gave them an environment where no one wanted to own technology stocks.HUMAN RESOURCES license revenue line marching to zero. Of course. During the shift. 211 .2 Concur P&L The Scrape strategy can work. Not something for the faint of heart. so this is not a change that’s easy to make. maybe we are entering another window of opportunity. Concur went from 700 employees to 300 and replaced the majority of the executive team. Figure 10. If you were to look at the stock price for Concur when CEO Steve Singh made the announcement that they would shift business models. but the transition will not be easy.

Those of you considering the acquisition strategy might consider the story of UpShot as told by founder Keith Raffel. In this strategy you’ll be best served by purchasing companies with products in adjacent markets. as anyone traveling through the San Francisco airport would have seen. Once acquired you’ll need to be cautious of the “white corpuscles”. In the late 90s UpShot was in head-to-head competition with Salesforce.CLOUD Buy New Buying a Model Six company is another strategy a Model One company can adopt. which will organizationally seek to envelop the foreign group and in time destroy it. Give them the opportunity to teach and share their experiences. Your challenge will be to leverage the knowledge of the people in development and delivery of the applications as well as in the new techniques for marketing and sales. Raffel has become an 212 . or make the same step in sales or marketing. Consider making their VP of Operations responsible for operations of the entire company. These days. or in with products that sell to customers significantly smaller than your current customer base.com. Perhaps the most famous example is Siebel’s acquisition of UpShot.

2008. Keith. Raffel was working at a start-up where there was no good way to send leads out to the field sales team.HUMAN RESOURCES author of a different kind. “HP might have been founded 31 Raffel.31 In the early 1990s. Raffel started by finding what he calls “a brilliant engineer. So Raffel came up with a way to distribute leads and send them out to the field using email and an electronic bulletin board (remember them?). The salespeople would then pipe up and claim that they already knew about any prospect the marketing team uncovered. the light bulb went on. a popular mystery set in Palo Alto. He showed the prototype to ten companies. Raffel took a leave of absence and started working on a prototype in 1995. and founded UpShot in 1996. When Keith saw the first Netscape browser in 1994. As Olson said. Smasher. and proceeded to rent space above a Jaguar dealer in Redwood City. having just released his second novel Smasher. who was between gigs” to build a prototype. and nine said they would buy it. but when the CEO asked him how effective the marketing programs were. Their aim was to make traditional CRM tools more accessible and affordable by using the Web. So he built a team that included Kris Olson to lead marketing and Bob Schulman as head of engineering. Midnight Ink. 213 . Not only that. he would point to all the deals that could be traced to one campaign or another.

I went to the hardware store. Siebel salespeople received their commissions stretched out over time. as we know. They signed up Fortune 500 companies like Johnson & Johnson. about the time Salesforce. unfortunately. Nevertheless. which was not as fully featured as UpShot. but UpShot was founded over one. Revenue for the UpShot product. From that point on. Xerox.” UpShot became the first CRM software delivered as a service in August 1999. and patched the walls. It took a year to fix this problem. is recognized over time and therefore. But ultimately they needed more money to fuel the sales and marketing engine (remember what we talked about in the last chapter). and HP. More momentum was lost.CLOUD in a garage. Siebel went with what they had developed internally. So. “A key engineer came to me one day and said that the paint fumes from below were leaking into his office.com was founded. bought some Spackle. in 2003.” Raffel tells a story from the early days and gives you a glimpse into the role of a start-up CEO. a big New York investment bank wanted to purchase a 214 . Siebel was ready to launch an internal Model Six product. In addition. they sold Upshot to Siebel. momentum slowed. at the time of the acquisition. And finally. unfortunately.

Finally. let’s focus on the organizational challenges you’ll have. at best to create new opportunities to package your service as part of a larger offering. there will be conflict on the delivery side with any premium support offerings or professional services offerings your company is providing. these new services will not endear you to any of your channel partners. everything changed again when Siebel was purchased by Oracle in 2006. These teams will say a standardized Model Four or Model Five service offering is too restrictive and worse yet will often block access to the customer. By the way. Your challenge will be to find was at worse to neutralize them.HUMAN RESOURCES huge number of seats of the Siebel Model Six offering and asked for new features. which will threaten them even more.” Remodel Finally. This brings us to sales. and still more momentum was lost. who not only resell your software. but also provide a box and a body. While we’ve discussed the advantages of doing this. First. As Raffel says. The decision to accommodate them. So it goes in Silicon Valley. made against the advice of the UpShot team. “That chapter is being written by others. While things improved gradually. took them even further off course. a well constructed Model Four service will be lower cost. Model One companies have the opportunity to create Model Four or Five businesses to run in parallel with their traditional business. 215 .

Oracle did not look to have any advantage. an SAP or Oracle application purchase price will be nearly the same. “In the end. So we said. based on the rule of thumb that the cost to manage the software is four times the purchase price. EVP of Latin America. over five years the cost of this application. and Alberto Chacin. VP for Oracle On Demand in Latin America. “In that case. probably somewhere near a million dollars. How? In a meeting with the mining company CIO we said. An application sales person can either double the size of the deal (application license plus first year of the service). devised a strategy that won the business. “What have you budgeted to manage the application?” We were guessing it was about $4M a year. Probably one of the best stories from the Oracle On Demand experience was competing head to head with SAP for an enterprise application sale to the largest mining company in Latin America. But Luis Meisler. or competitively out position other traditional software providers.” The CIO agreed.CLOUD Your biggest advantage in the Remodel strategy is that you can leverage the existing sales force. whether from SAP or Oracle will cost you $31M: $1M to purchase the 216 . as SAP was both stronger in Latin America and stronger in mining as a vertical market. The CIO answered $6M per year.

you’ll have to create an ondemand sales overlay team and focus on delivering a lot of information about how this solution is simpler.” The Oracle sales person assigned to the mining company not only won the business from SAP but also received a significantly higher compensation.HUMAN RESOURCES software and $30M to manage the software. is part of a growing number of MMORPGs (Massively Multi-player Online Role Playing 217 . cheaper. and better than doing it the old-fashioned way. you can buy the software from SAP and manage it yourself or you can buy the software from Oracle and manage it yourself. there is a third alternative: let Oracle manage the software for $15M. given that the deal included the license as well as the first year of the services. So. Once you have that core group of reference customers. Organizationally. an elaborate fantasy computer game. once your application sales force is compensated for selling On Demand. you’ll be ready to ‘cross the chasm’. you will quickly bring them into alignment with your goals. In either case it will be $31M. World of Warcraft. World of Work So how much will our work lives change in the future? What will be the nature of work? Here there are lessons to be learned from the world of online gaming. So you can see. However.

how is it that World of Warcraft accomplishes something that we find so difficult in the world of work? One of the keys to the success of World of Warcraft is the idea that an individual's skills are known and non-overlapping. at any moment in time. In the 787. versus a previous mark of 50 percent. Sounds like the modern World of Work. These suppliers span the globe as shown below. there are hundreds of thousands. some young. With over seven million subscribers. outside suppliers are responsible for about 70 percent of the aircraft parts. Some characters have the ability to kill. some old. but not heal. people they don’t. if not a million. What are they doing? While some are merely spectators. the 787 Dreamliner. 218 . The company has pushed outsourcing of component parts to new levels. doesn’t it? So. in London—all to achieve an objective. Consider the latest Boeing aircraft. And other characters can heal. the vast majority are working with other people—people they know. people simultaneously playing the game.CLOUD Games). in Beijing. Our business world needs to extend these fundamental ideas in software that brings information and people into a network. but cannot slay.

Leighton Read. 219 .HUMAN RESOURCES Figure 10. Europe. led a team that surveyed CIOs regarding the SaaS model. They learned that the proportion of CIOs considering adopting application cloud services in the coming year has grown from 38 percent the 32 Byron Reeves and J. and North America to talk. Summary Early in 2007. Leighton Read: Games at Work: How Games and Virtual Worlds Are Changing the Way People Work and Businesses Compete. but our software isn’t. If you want to read an entire book on the implications of games check out the new book32 by Byron Reeves and J. Business managers can no longer ‘manage by walking around’. Today we’re all doing conference calls at 4:00 in the afternoon. because it’s the best time for Asia. Abhijit Dubey of McKinsey & Co. like the great David Packard taught a generation of HP managers to do.3 The World is Flat The world might be flat.

Dubey and Dilip Wagle. they’ll have to build the management processes and organizational structures to manage two distinct but potentially mutually cannibalizing businesses (traditional software and software as a service) with different business models and requiring remarkably different sets of capabilities. have published a report entitled “Delivering Software as a Service. 220 . software executives who continue to put it off risk being left behind. Although this challenge is formidable.” There are many interesting comments in the report.” We could not agree more. but perhaps the best quote is: “And finally. a partner at McKinsey’s Seattle office.CLOUD previous year to 61 percent in 2007.

you’ve found it enlightening and entertaining. Word.BEGINNING OF SOFTWARE Congratulations on making it to the end of the book. Powerpoint. 221 . this is the third wave of computing. following mainframes and client-server then we are indeed at the beginning of another generation of software as revolutionary as Excel. If. as some people believe. SAP Financial applications and Oracle databases were in the last generation. and you’ll repeat some of the campfire stories. Hopefully.

I told him it has to become more about the information and less about the software. A few years ago. most of my customers are starting to ask me—what is the best sales comp plan?” 222 . and he called me again and said he’d made the decision and now that he was well on his way. Of course. Do you attempt to build a Model Four business? Do you acquire a Model Six business. or do three of your colleagues leave and start a new venture using your specialized expertise? For those of you already in a Model Six business. you have many choices. Time passed. perhaps the book has given you a better understanding of the challenges your fellow team members face. I told him all of the good reasons he should consider doing it. “Interesting. the CEO of a company that builds sales compensation software in the traditional model asked me if he should deliver his software as a service. He said. he asked me what he should think about next. and you’re starting to think about how information and community will change your business in the future. near infinite storage and a network of mobile eyes and ears that number in the billions. For those of you in the traditional software business model.CLOUD This software can take advantage of 1000s of computers.

S. but here the customers are asking for something even more valuable: knowledge. Concur today processes over 10 percent of the expense reports in the U. the best seller. the top 10 places to stay in Tahiti. information. We already have pointed out that consumer Internet applications are all content rich. eBay or Google as software companies. because what we see when we when we use their applications is the information: the Pez dispenser. and what times of the year travel is happening is far more valuable than the transactional efficiency afforded by delivering expense management as a service. Perhaps this is why few of us see Amazon. The information about what hotels. All of you who are running software as a cloud service are sitting on valuable business knowledge.BEGINNING OF SOFTWARE Now most of us in the software business are used to being asked for more features and functions. which airlines. What is the value of knowing where there is a deficit in skills. by geography? What if an ondemand CRM software vendor were to analyze all of the configured workflows and identify patterns of sales processes with the shortest sales closing periods? Again. Taleo’s systems have seen millions of resumes. what VP of Sales would not pay to know what sales processes are the most efficient in a particular industry or geography? Few people may 223 .

as many will be nervous about their trade secrets being published as best practices. what rules apply to aggregated information? When you log on to Amazon. the United States’ government contracts with Visa and MasterCard to understand aggregated data of how much each of us. 224 .CLOUD realize this. whether you’re part of a large software company or three people with a great idea in Olathe. Kansas. But. such usage of customer data remains a controversial topic.” We hope this book has given you some insights into how to think about the future of your business. the value of the information far outweighs the value of the credit card transaction processing. We’re much closer to the Beginning of Software. spends. Publication and resale of identifiable private data obviously violates legal regulations and privacy laws. Of course. or what some call the “Golden Age of Software. but to assess the overall health of the economy. Again. as a consumer. but hopefully you get the point. Software is far from a mature industry. Why isn’t this the case for business software? We could go on and on with new companies and ideas. you’re contributing to their overall knowledge of consumer interests with every click.

225 .

226 application cloud service companies · 28. 180.INDEX A Acquisition strategy to transition from Model One · 212-213 Amazon · 17. Ed · 148 Beasly. 154 search-based · 154 usage spectrum · 100 weight · 98-9 Aspiren · 61 AT&T · 71-2 availability · 81–83 Azure · 111 management · 69-70. 116 Bezos. Fred · 104-6 Brooks’ Law · 105 Brown. 95 applications frequency of change · 99 SQL-based · 152. 208. 87-8. 142-5. Jeff · 75. 181 management at eBay · 139-141 operator errors · 144 Black. 182 Blueroads · 188 Blum. Marc · 42-3. Peter · 62 Blackbaud · 52-3. 226 Animoto · 87. 82 power footprint · 67-8 reduction · 124 B Back of the Box Network · 59 backup power generators · 63-4 Barrett. 77 billing · 198 226 . 223-4. 42-3. 75-8. Chris · 25 carbon computer footprint · 91-95 credits · 218 emissions · 67-8 impact · 64. 129. 153. 226 Apex · 114. Gloria · 29 Bruck. 110. Chris · 30 Bootstrapping Your Business: Start and Grow a Successful Company with Almost No Money (Gianforte) · 46 Brooks. 72. 168. 31. 34. 179. Boyd · 47 Benioff. Michael · 150 business models traditional (Model One) · 3-5 open source (Model Two) · 5-7 outsourcing (Model Three) · 7-9 hybrid (Model Four) · 9-14 hybrid+ (Model Five) · 14-15 SaaS (Model Five) · 15-17 Internet (Model Seven) · 17-18 C CAL (central application logging) · 141 Capability Maturity Model (CMM) · 106 Capossela. 184.

28-30. 196-7 Cisco · 27 Clementi.INDEX CEMLI (Configuration. 128-9. 126-7 Constant Contact · 40-2 consumer Internet applications · 18 companies · 17 usage spectrum· 100 contracts · 178–79 corporate culture · 208-10 cost of software· 18-22 Crossing the Chasm (Moore) · 204 customer service · 4. Bibby) · 187 End of Software. Gelman. 99. and Integration) · 11. 223 profit and loss · 211 Configuration. 124. 102. Localization. 160. 168. 58-9. Larry · 30. 13941. 161. 102. 22. Extension. 104. 43 email marketing · iii. Lars · 35-6 data center facilities availability · 69-70 disaster recovery · 58 power carbon footprint · 67-8 power quality · 63-4 227 . Marc · 180. 19-20. 182 churn · 190. Alberto · 216 change management · 4. 79. 198 Power Utilization Efficiency (PUE) 64-5 database architecture · 131 DealerTrack · 51-2 deep Web · 151-2. 126-7 CEMLI framework · 126 Chacin. Modification. Cooke) ·138 Estimating Total Power Consumption by Servers in the U. 228 Emerging SaaS Channel. 182. and the World (Koomey) · 67 eTapestry · 53 evaluation criteria. 47. Mark · 83 Ellison. Abhijit · 220 E eBay · 17. Dilip) · 219 disaster recovery · 12. 36. Erich · 84 cloud computing · xxx Cloud Computing Test Bed · 90 CloudFront · 78 CMM (Capability Maturity Model) · 106 Computer 2000 · 188 Concur · 16. SaaS vs. 210-1. Localization and Integration (CEMLI) · 11. Modification. The (TenWolde. 124. Extension. 45. 228. 79. 154 Delivering Software as a Service (Dubey. 37. 226. financial · 200 D Dalgaard. 137-142 channels indirect · 186-9 Chardon. 228 Ellis. 228 economics of software. Matt · 78 Dot Dead (Raffel) · 212 Dubey. 223. The (Chou) · ix EnergySys · 61 ERP II: best practices for successfully implementing an ERP upgrade (ACM) ·138 ERP Trends (Peterson. 150. 200. 109. 232-3 availability management · 141 change management · 140 traditional software cost example · 18-19 EC2 (Elastic Compute Cloud) · 75-7. 104. 104 distribution channels · 186-9 Domo. traditional · 18-20 EDS · 7 Elastic Compute Cloud (EC2) · 75-7. 194.S.

219 Hiding in Plain Sight: Google Seeks an Expansion of Power (Markoff. 113. 213-4. The (Shih) · 165 facilities. 205. 26. 6. Pat · 121 Gates. 93. 17. 228 Goldberg. 73. Joe · 174 Greenstein. Michael · 124. Guido · 121 Haig. Greg · 45-7. Henry · 125 free software movement · 5 Front of the Box Network · 59 I IBM · 5. Ed · 166 Hewlett-Packard (HP) · 90-91. Johnson. 219 Hsieh. 213-4. The (Moore. 163-4. Andrew · 80 Graves. 123. 228 Ford. 8-9. The (Gates) · 110 Iyar. Subrah · vi. 117-19. 229 Kenexa · x. 87. 65. Tom · 206 Knol · 153 Konary. 229 Innovator’s Dilemma · 203 Internet Tidal Wave. Hansell) · 93 Hilson. 228 Gibbs. 5. 84. 128 J James. 26. 89. 208 Hummer Winblad · 201 F Facebook · xiii. Kevin · 111 Gluecode Software · 5 Goldberg.34. 226. 188. 26 G Garrehy. 17. Josh · 39-40. 229 Global Services · 7-8. Evan · 129-31. 49-50. Bruce · 34 Held. 2293 JasperSoft · 6 JBoss · 6 Jeff Bezos' Risky Bet (Hoff) · 75 K Karsan. 204. data center availability · 69-70 disaster recovery · 58 power carbon footprint · 67-8 power quality · 63-4 Power Utilization Efficiency (PUE) 64-5 financial evaluation criteria · 200 Flock IT · 49-50 Force. Don · 133 Haley. 223. 151-3. 228 Facebook Era. 128 HP · 90-91. Jay · 36 Gregoire. Rudy · 32. Kippola) · 206 Gough. Ken · 47 Hatz. 195. 91. Jerry · 79 Herrmann. 229 King. 228-9 Gianforte. Mike · 80 228 . 31-3.com · 113-14. 182 LotusLive · 26 Rational · 86 indirect channels · 186-7 Infosys · 8. Tony · vi. Amy · 3 H Haarmans. Scott · 161 Hilton. 163.CLOUD Harris. 228-9 App Engine · 111-3 Bigtable · 112 Gorilla Game. Bill · 110. Joe · 172 Kippola. Randy · 34 Goodman. Michael · vi. Gail · 40-1 Google · xiv.

Glenn · xxx Linux · 6. 152. Geoffrey · 204 Morgan. 76. 28. Brenda · 36 Lederman. 222 Model Three. Hybrid+ · 14-5. 232 NTT · 61-2 Nyirjesy Bragale. Software as a Service · 9. 2300 Nippon Telephone and Telegraph (NTT) · 61-2 No Silver Bullet: Essence and Accidents of Software Engineering (Brooks) · 150 Nordstrom’s guarantee · 10. 191. 214-5. 4. 230. 52. Luis · 216 metrics · 64. 15-7. 120. 208. 99. Dave · 30 license model. 202. 186. 210. 222 Model One. Fred · 127 Manber. 178-9 Model Two. 53. Open Source · 5-6 Monthly Recurring Revenue (MRR) · xiv. N Nazir. 172. 54. 230 Miller. 22. 210. Internet · 11. 193-4. Bill · 41 MMORPG (Massively Multi-player Online Role Playing Games) · 217. 180-1. 142. 54. 186. 186. Laura · 200 Leighton. 65 NetSuite · x. 215 Model Seven. 12. 123. 129. 11-2. 22. 191. traditional · 9. Bob · 37 Lervick. 226. 83 Model Six. 195-6. Hybrid · 9-16. 122. 212. 16. 80. Zach · v. 193-8. 175. 52. 162. 171-5. 62. 210. Read· 219 Leingang. 81. Priscilla · 192 Morin. 101. 188. Mark · vi. 191-2. 210. Bob · 115 MuleSource · 29 MySQL · 123 Mythical Man Month (Brooks) · 104 L Leadley. 188. 30. 193 Moore. Christine · 38 O O’Neil. 94. 122. 14. Udi · 153 marketing cost of · 17 email · 160-1 expenses · 4. 210. 52 Olson. 30. Hank · 178 Lentz. 232 Microsoft · 25-6. 230 McCoy. 6. 131. 212. 197. 230 Model Five. 22. Traditional · 3. xiii. 17. 185-6. Jonathan · 95. 179. Dave · 117 Morrell. 129. 9. 120 Netflix · 64 NetLedger · 21–22 Netsuite · 21–22. Siara · 161 Nelson. 180-2. 20. 13. 191. 30. 230. 72. 229 Magner. 210 LotusLive · 26 LucidEra · 170 M Magic Number · 190. 232-3 search marketing · 162-3 Massively Multi-player Online Role Playing Games (MMORPG) · 217. 76 Lipscomb. 46. Kris · 213 229 . Outsourcing · 7-10. Courtney · 29 Meisler. 228-9 171. 173-4. 171. John · 153 Lim. 30-1.INDEX Koomey. 230 Model Four. 190. 215. 215. 113. 109-10.

53Openwater Networks · 154 operator errors · 144 Oracle · v. 17. 230 applications · 11. Ray · 110 Q QA · See software:Quality Assurance Qlusters · 29 Quan. 195. 73. 99. 27. 4. 105. 115. David · 219 Palmisano. Aaron · 176-7. 43. Russia · 9 India. 49. or a Tool to Increase Value? (TPI) · 178 revenue · x. 49. 28-9. 134. cost of · 63 Powers. 192. 204 sales · 15 business · 61 Perry. 216 OSI · See Open Source Initiative Ouellet. 67. ix. 32. 93-5. 12. 34. 93. 126-7. Katy · 162 Rudin. 231 Roth. 81. 128. Keith · 212-5. 187-8. 109. 231 recruiting · 11. Everette · 30 power · 33. Byron · 219 reliability · 57. 69. The · 208 sales channel · 13. 171 compensation · ii. 231-3 RightNow Technologies · x. Dennis · 130 R Raffel. 10-12. 84. 207 Reeves. 231 Reedy. 39-40. 215-17. 65. 172. 83. 40. Ken · 170 Rudman. 31. 19. 204 Recruitsoft · 33 Red Hat · 6-7. 7-9. xiv. 92 S S3 (Simple Storage Service) · 94–99 SaaS · See Software as a Service SABRE Group. 144 Public Data Sets ·78 PUE · 65-6. Rick · 37 P Packard. 125 Riskonnect · 115 Rocha. 22. 185 data center · 149 maintenance · 3 product support · 154 Oracle On Demand · 10. 180. Sam · 84 performance management · 136. 51. 66-7. 222 230 . 138-9. 227-9 power consumption · 64. 190-2. 59. John · 38 Phillips. 228. 57. 77. 38. 14. 49. 231 risk management · 88. 32. 63-4. 191-3. 89. 182-3. 149. 134. 79. Lynn · 139. 145. Todd · 164 Pestana. 230 Open Source Initiative (OSI) · 5 open source software · 5-6 OpenTable · x. Dave · 80 programmers · 5. 85. 46. Michael · 154 Ross. Martin · 33 outsourcing Brazil.CLOUD Omniture · x. Eastern Europe · 9 Ozzie. 125. 182. 117. 125. 142. 201. 141 remodel strategy to transition from Model One · 215-6 resellers · 186-7 Restructuring Outsourcing Agreements: An Indication of Failure. 3. 95 power outages. 103. 133-4. 45-7. 214 revenue recognition · iii. China. 55. 25.

economics 21. Marian · 125 T Taleo · x. 173. 232-3 Schulman. 179. 191. 33. 34. financial · 100. 31. 51. 95 technology powered companies · 91 traditional software companies · 13. 182. 91. 95. 210. 82. 118-19. 153. Tien · 193. 197-8. 233 technology birth year · 92. 1. 164-5 social networks· 163-4. 4. 232 Nordstrom’s guarantee · 180-1 Service Science · 9. 74. 232-3 UpShot · 16. 172. 181 Mexico treasury secretary example · 149 Security Threats from Within (Bruck) · 150 Service Level Agreements (SLA) · iii. 32. 103. 99. 114. 71. Bob · 213 Schultze. 167. 170. 211 SLA · See Service Level Agreements Smasher (Raffel) · 213 social graph· 89. 42-3. 133. 212. 176 teams · xiii. 199. 227-8. 207. 210-11. 22. xiii-xiv. 212 SAP · 3. 212. 145. 175.43. 100. 148-50. 97. 232 Seven business models · xi. 125. 43. 55. 216-17. 204. 57. 85-6. 220 sofware companies · xv. 232 Care Rehab example · 148 management · ii-iii. 1617. Axel · 188 Scrape. 187. 153-5 search · 49. 131. 149. 15. 203. 27-8.INDEX productivity · 157 sales force automation (SFA) · 49. 35-6. 152. 14. 160. 21. 223 social networking · 146–48 software as a service (SaaS) · viii. 80. 132. 146-7. 137-8. 169 software business ·xiv. 198 cost of · 18-22. 226. 173. 201. 99. 200-1. 179. 2 Singh. 170. 198. 203–6 Spohrer. 103. 164. 191. 4-7. 128. 113. 48-9. 107 subscription · xiv. 103. 226. 10. 214. 16. 231 applications. 181. 155. 19. 16. 21. 233 usage spectrum · 100. 32. 9. 174. 56. 29. 13. 137. strategy to transition from Model One · 210-1 search-based applications optimization · iii.com ·x. 31. 181. 182 security · xii. 95. 232-3 standardization · 11-12. 97. 134. 233 transitioning from Model One · 29-30. 34. 197. 155. Jim · 9 Successfactors · 31. 70) · 82-3. 123. 30. 233 231 . 221 SAS 70 (Statement on Auditing Standards No. 101. 198 U Udzinski. 39. 49. 230. 228-9 Twitter · 163-4 Tzuo. 151. 202 specialization · 53. 83-4. 137. 86. 226. Chuck · 37 upgrades · 4. 174-6. 171. Steve · 29.228 traditional · xi. 180. 77. 176-7 profit and loss · 201 salesforce. 233 storage cloud services · xii. 169. xxi. 70. 204 Szefler. 231-2 specialization · 12. 226. 11-12. 198 productivity · 42. 70. 76. 230. 148-9. 173-4. 180.

151. 80-2. 200. 157. 16. 201-2 World of Warcraft · 15. 37-8. 153. 111 Visualforce ·114. 32. 204. 234 Z Zaidenweber. 200-1. Dilip · 220 Wainewright. 234 Vocus · x. xiv. 234 Vertica Systems · 97 Viasite · 11 virtual machines · ii. 217-18. 76. 234 V value added resellers (VAR) · 187. 25-7. Phil · 114 WebEx · x. Ann · xiv. 234 W Wagle. 234 Vogel. 233 vendor-specific objective evidence (VSOE) · 191-2. Werner · vi X XenSource · 29 Y Yahoo · 17. 174-5. Natan · 116 Zappos · 209-10 Zipcar · 192. 234 232 . 163. 182. 33.CLOUD profit and loss · 169–70 Winblad.

Sign up to vote on this title
UsefulNot useful