Professional Documents
Culture Documents
An Introduction To SAP HANA PDF
An Introduction To SAP HANA PDF
An Inttroduc
ction to
t SAP
P HAN
NA by Dr.
D Vis
shal Sikka
S
UNIT 1: Intrroduction and Backgro
ound of SAP
P HANA
00:02:13 It was milllions of times ance than it is now. Also,, the other fundamental
s worse in prrice/performa
thing was that CPUs backb then weere single core.
00:02:32 And already by 2003 it was clear tthat this was running into a physical w
wall, into a ha
ard
limitation,
00:02:48 in order to
o continue the performan nce and manuufacturing be
enefits of Mooore's Law. So
S already
by 2003 also,
a it was cllear that a co
ompletely ne
ew kind of database paraddigm was within our
reach.
00:03:03 And whenn we started thinking abo ut this, it bec
came clear thhat if SAP haas to build a new
database,, then it has to
t be built arround the new
w reality of hardware:
h
00:03:15 That is mu essing; masssively larger and cheaper main memoory; and the advent of
ulticore proce
columnar structures.
00:04:48 So, when we looked at nking about this back in 22002. My first conclusion
a this, when I started thin
was that SAP
S had to build
b a new d
database.
00:06:13 back in 20
003. When Franz
F showed
d the EUCLID
D running a billion recordds in one sec
cond, that
was prettyy remarkable
e.
2
00:06:25 So Hasso o started to te
each these thhings at HPI around 20044, and startedd to launch a real
investigation into how all of these tthings could be combined into a full ddatabase.
00:06:47 And he wa
anted to rew write Financia
als for the fou
urth time in his life. And thhis time, bec
cause of this
amazing ability
a to calc
culate things on the fly, and the drama atic performaance that we e get
00:07:46 And so we
e started worrking on thatt. In 2009, Ha
asso presentted his pape r on the in-m
memory
database,, this column
nar in-memorry database,
00:07:59 at SIGMO
OD in New Enngland. And it was extrem
mely well rec
ceived. That Fall, we starrted the
HANA devvelopment project
00:08:36 So HANAA became gen nerally availa ortly after Sappphire, back
able in June of 2011, sho k then. At
Sapphire, I had showe
ed the first 25
5 or so custo
omers that we had workeed with
00:09:09 As we are
e taping this it is Septembber of 2013, so it's about 2 years andd 3 months siince the
launch of HANA. Actually, it's exacctly 2 years and
a 3 months s! It was Junne 20th of 2011 that
HANA beccame genera ally available
e.
3
00:09:54 And it hass just been a hell of a jou rney since th
hen. Beyond the customeers, 10 different
hardware vendors are e manufacturring hardware e for HANA.
00:10:19 It is now 10
1 years sinc
ce we started
d working witth Intel. I rem
member whenn the first Wo
oodcrest
chip came e out,
00:10:38 We just reecompiled the software fo or the Woodc crest chip. Actually, it waas the same code.
c Then
it was clea
ar that this multicore
m ben
nefit that we get
g from HAN NA
00:10:50 was some ething that was going to rreally... So, thanks to ourr friends at Inntel. You are the best!
Without yo
ou, HANA would not be p possible.
00:11:54 or is in the
e process of running on H HANA. And one
o of the most amazing things that our
o team did
recently, that
t we are really, really p
proud of,
00:12:03 is ISP. Ou
ur internal ER
RP system n ow runs on HANA.
H And this
t has basiccally more th
han 60,000
users in our
o company.
4
UNIT 2A: S
SAP HANA Technology:
T Parallelism
m
00:00:00 So let's ta
alk some HAN gy. And yes, every once in a while, w
NA technolog we have to ta
alk
technolog gy.
00:00:16 We can't be
b all about PowerPoint and traffic lig
ghts and things of this soort all the time
e.
00:01:54 2 terabyte
es of DRAM, and you can
n put maybe 5 or even more terabytees of SSD as the
persistencce on the serrver.
00:02:18 e assumption
Unlike the ns of the passt, where peo ople used to try
t to keep thhe CPU cons
sumption
low, and so
s forth, here
e, our belief iis the more you
y burn the CPUs the faaster you gett the results,
00:02:51 believable am
It's an unb mount of com
mputing capaacity. And you have 2 teraabytes of data in DRAM.
So everything on HAN NA was desig
gned to take maximum ad dvantage of these two thhings.
5
Seattle,
00:03:59 anything of
o that sort: If it requires llet's say 350 billion scans
s, you can doo this in 100 seconds on
one core,
00:04:07 or more or
o less in one
e second on a hundred co ores. This is basically whhere the power is derived
from. We also do in ad
ddition to the
ese scans,
00:04:15 because of
o the native algorithm in the operators, we do about 12.5 to 115 million agg
gregations
per secon
nd per core.
00:04:54 What thatt means is that basically tthat not only do we take a little job annd distribute that across
processorrs, we can evven take, witthin the operaator itself, we
e can take a part of the jo
ob
00:05:30 So that is: Number one is the para allel operatorrs. Let me seee... where iss the? Yeah, there is
ay has this fu
this, Sanja unky icon tha
at sort of looks like that
6
UNIT 2B: S
SAP HANA Technology:
T Row and C
Column Storres
00:01:08 And so we
e do latch-fre
ee index travversal. So this was a new
w data structuure that Sang
g and his
team had designed, and so on and d so forth.
00:01:14 So we havve both of these. And thee benefits, off course, of the row storee are that you
u can do
transactio
ons quite quic
ckly. The ben
nefit of the co
olumn store, like I alreaddy talked aboout,
00:01:41 one of our sales orderrs, for examp anufacturing order, or thee account se
ple, or the ma egments, the
BSEG,
00:02:20 So it is a very,
v what we
w would call wide data sttructure. And
d when you hhave such wiide data
structuress and whenev ver someboddy, a normal human being,
00:02:43 So in a co
olumnar dataa structure, th
hat means thhat you just pick
p out the oones that you u need, that
you are in
nterested in getting
g inform
mation on, qu
uickly assemble them intoo the result,
00:02:54 and demo ore, in a tradiitional disk-based row stoore, you are going to the
onstrate that.. In a row sto
disk, you are grabbing
g things row by row, and then after yo ou have retrieeved all the rows,
r
00:03:08 Here, not only do you get just the ccolumns that you are inte
erested in, buut, in fact, do
o that
massivelyy parallel, bec
cause you caan assign diffferent cores to grab the different coluumns.
00:03:19 In fact, yo
ou can take more
m than on
ne core for on
ne particular column. Andd this is the fancy
f
7
cocktail th
hing that I told you about,, the intra-op
perator parallelism.
00:03:48 So you haave the basic c column sto re here. In addition, we have
h a less ooptimized collumn store
that we ca
all the delta store.
s
00:04:40 is we have
e added a co
oncept of wh
hat we call an
n L1 delta, which is a variiation on this
s row store,
00:05:56 and absorrb the transa actions at a vvery high spe eed into the row
r store. Annd when you think about
it, the worrld sort of wo
orks like that.. If you ever go
g to China, and you havve a translato
or who is
sitting the
ere and translating things for you,
8
main inve
ention, and in
n addition the
e parallel row
w store
00:06:43 And keep in mind: One of the thing gs that people get confus
sed about in HANA is thaat people
e is for transsactions and the column store
think that the row store s is for aanalytics:
9
UNIT 2C: S
SAP HANA Technology:
T Projections
s, Dynamic Aggregation, Integrated
d Compress
sion
00:00:56 Like that, and some off these thing s are filled, and
a then you
u do from theere into just those two
that you grabbed.
g Like
e that.
00:01:38 when you need to get, let's say, alll the custom
mers which arre overdue, aand their add
dresses, that
is maybe 10 fields outt of these 320
0.
00:02:42 In the worrld of HANA this is not neecessary, because of the tremendouss scan speed
d that I
mentioned d; three and a half billion scans per second
s per co
ore.
10
transactio
onal data,
00:05:14 00 countriess in the world, so if one off these fieldss is Country, you know
there are like, what, 20
that there are only 2000 values out of there.
00:05:30 et a dramatic
So you ge c compressio
on improvement without compromisin
c g performance. And that
is something quite am
mazing about HANA,
00:06:49 And out of s the actual database sizee, and the remaining
o this, sometthing like 1.1 terabytes is
700 gigab
bytes is the working
w mem mory of HANA A.
11
00:06:59 So this is quite amazin
ng that we arre able to ge et that much compression
c n, and as a re
esult, I
mean, if you
y think aboout it, if you g
get rid of the aggregates,
00:07:39 So that wa
as projection
ns, dynamic a
aggregation,, and the inte
egrated comppression of HANA.
H
12
UNIT 2D: S
SAP HANA Technology:
T Insert Only
y, Partitionin Out, Active aand Passive
ng & Scale-O e Storage
00:00:43 ve and passiive storage. So, INSERT ONLY is, whhen we have
and hot and cold, activ e a column
store,
00:01:02 And so yo
ou add new columns
c in th
here. Adding a new entryy in here meaans taking a part of that
record here and just making
m the rig
ght insert in the appropria
ate place.
00:01:25 then in co
olumn storage
e, it is very a
advantageous to simply create
c a new
w entry, and then as a
separate process, inva
alidate the p revious entry
y that you ha
ad.
00:01:55 And in HA
ANA, we get this natively.. In fact, the update sequ
uence operattion is implem
mented in
HANA
00:02:04 bination of an
as a comb n insert and then an inva
alidation of th
he thing that'ss not valid an
nymore.
00:02:37 ...maybe he
h drew it likke that, and t hen this has four pieces, and then thiis one is like
e a zillion
pieces, an
nd stuff like that.
13
00:03:33 or multiple nd of course,, you can partition columns, so you h ave one piec
e servers. An ce of the
database in one mach hine, and so on, or a com
mbination of both
b of thesee things.
00:03:42 And then unleash the cores that a re sitting on each one of these mach ines into all of
o them.
wesome perfformance.
Then you get really aw
00:05:00 2 trillion, in ca
That's 1.2 at's 1.2 and then 12 zeross after that.
ase you are ccounting, tha
00:06:41 as well ass the items from this yearr that you will carry forward into the neext year: So maybe 14
months off information for one yearr. And then iff you want too do year oveer year comp parison,
00:06:49 so then yo
ou'll do also the
t previouss year. So tha
at's it. You do
on't need to hold more th
han that in
14
the hot... in active mem
mory.
00:07:17 so that we
e can get eve
en more commpression, th
hen we can getg even betteer performan
nce. And
HANA ena ables us to do
d these kind
ds of things natively,
n insid
de the databaase.
00:07:28 So that wa
as INSERT ONLY,
O partit ioning, and scale-out,
s and hot and coold.
15
UNIT 2E: SAP HANA Technology:
T SQL, Libra
aries, and Su
ummary
00:02:10 I mean ba
asically, when you think a
about it, a ma
ap is like a sc
can that you distribute ov
ver a large
amount off data when you want to map an operation,
00:02:51 And then we have the stored proce age SQLScript. We havee a native, low
edure langua w-level
language called L, which is a part of the LLVM
M,
16
Predictive
e Analytics Liibrary.
00:03:20 We integrrate R, which h is a statisticcal package in there, we have IMSL.. ., let's see... all kinds of
function libraries like this,
00:04:40 working on
o any databa ase that you can think off in the world, you can goo and, withou ut any
training whatsoever,
w get
g up to spe ed and running on HANA A. It's fully staandards-com
mpliant.
00:05:52 and other stores that wew can add llater on, as we
w think about it. And texxt we use the
e column
store for. So this is als
so text.
00:06:02 And then we have bey yond the core e data types and stores, we have thee engines tha
at work on
these. So these are th
hings like the
e OLAP engin ne,
17
more engines here as we go. The graph engin
ne will come here as well..
00:06:41 And so on
n. So there are all these e
engines that are inside HANA.
H
00:07:21 And then we have all kinds of libra e. And these libraries are,, I already mentioned,
aries up here
these are things like BFL,
B
18
UNIT 3: SA
AP HANA Pe
erformance Benchmarks
B s
00:01:25 A couple of
o years ago, Fujisawa sa an told me th
hat he is the head of IT T and operations at
Yodobashhi, and he's also
a the son o
of the ownerr of the firm. He's from thee founding fa
amily
00:01:51 So that's 22
2 million. And out of tho
ose, 5 million are loyalty. So this is thee total. And what
w they
do is, theyy used to do in our ERP ssystem on an Oracle database,
19
that has come
c in. It's a very compllex query.
00:04:12 So how do
o we think abbout this in a more scientific way? We
ell, my sensee is that in th
he age of
HANA, wee have to rethink the con cept of perfo
ormance itself,
00:06:34 the more of these five dimensions are in there, the more HANA's perforrmance stan
nds out.
20
u can imagin e.
know, as many as you
21
UNIT 4: SA
AP HANA Ro
oadmap and Re-thinking
g Software Developmen
D nt
00:01:47 e bringing HA
So we are ANA to everyy single product, every single applicattion that we have.
Beyond th
hat, all the technology pro
oducts.
00:02:19 that we co
ould non-disrruptively put HANA underneath BW, butb in a way that dramatically
accelerate
ed it. So the content coulld remain unchanged. It got
g compiledd into HANA.
00:02:41 And the looading time into BW has been made parallel, so parallel
p loadinng. That mea
ans things
like the DS
SO activationns and the P
PSA activatio
ons. These arre the staginng areas.
22
ed, BW le
picture. Evverything here I mentione et me change
e the color off this guy to white,
w
00:04:09 And the Java platform n HANA. So the new HANA Cloud plaatform is run
m runs also on nning on
HANA.
00:04:31 This is whhat we call the XS engine e. This is the native applic
cation servicces inside HA
ANA. So the
three tier plan was kinnd of invented
d as an efficiiency mecha anism.
00:05:46 and that iss something that is quite amazing. So o basically, when
w you thinnk about HA ANA, the
application platform sttory around tthis is that we
e have three e categories oof application
ns that run
on top.
00:06:40 And of course with thee native, we h wnership thatt you can run
have the least cost of ow n the entire
application inside of HANA.
H
23
and its evvolution, one amazing thin
ng that we ha
ave done, that our teamss have done,
00:08:08 and that iss just awesome. And so, one great th
hing there ha
as been that w
we have bee
en able to
eliminate batch jobs and replace th
hose by realttime.
00:08:58 And todayy, software development means you have h people writing codee; testing is an
a offline
process; there
t are tea
ams that are distributed, they
t cannot collaborate
c wwith each oth her in
realtime;
00:09:38 e browser as
both in the s well as in E
Eclipse. And the
t team in Israel, Jake aand Ariel and
d the team
are workinng on River
00:09:47 as a way to
t rethink an
nd simplify th e developme
ent experienc
ce with instaant feedback,,
24
responsivveness, the ability
a to test code inline on
o the fly,
25
UNIT 5: SA
AP HANA in Practice and
d Summary
y
00:01:06 Route calculations, forecasting of utilization of containers; these are thee kinds of problems
e can run thin
where we ngs tens of th
housands of times faster.
00:01:33 and using hat the CFO of eBay callss the eBay economy.
g signals to figure out the health of wh
There are
e 300 people, analysts, inn eBay who work
w on analyzing signalss,
00:02:36 on running
g the gene trreatment thin
ng, and align
nment, and va
ariant callingg, and so fortth, dozens
to hundreds of times faster
f with HA
ANA.
26
00:03:20 In the oil industry, for example,
e witth the exploration of oil, we
w can do seeismic data analysis
a for
exploratorry work that big oil compa anies do.
00:03:41 Every timee a drill gets stuck, there is tens of millions of dollars of expennses that hav
ve to be
incurred to
o remove the e drill,
00:04:02 And this iss something that is quite extraordinarry. So if you are thinking about this, I think just
think abouut something
g that is desirrable, feasible, and viable
e for custom ers.
00:05:27 So refacto
oring, and rethinking, and
d re-imaginin
ng these is so
omething thaat comes natturally to us.
But there is a ton of am
mazing thing
gs that can be done with this technoloogy.
00:05:50 And so we
e've been thiinking about new areas, for example,, things that w
we did like HANA
H One,
00:05:59 which is th
he deploymeent of HANA that is availa
able on AWSS, but also onn Korea Tele
ecom, on
Portugal Telecom,
T and
d many othe rs; on VMwa are, and so on.
o
27
00:06.46 But, with the
t efficiency y of HANA, t he efficiency
y of the Cloud
d, and the abbility to do elastic
deployme ents of large scales,
s
00:07:15 Accenture
e has done a great set off things arounnd retail and thinking aboout consume er proximity,
customer segmentatio
on, and thinggs like this. Deloitte
D has been
b buildingg applications
s.
00:09:35 One day I got a mail frrom the CEO O of Taodynee that they arre doing visuualizations off stars using
HANA: Annd of course,, the first rea
action was, what
w the hell is that?
28
00:10:12 and you can,
c you knoww, fly around
d at warp spe
eeds through h the universee. And the challenge is
that when
n you look at the universe
e from a particular angle,
00:10:45 Or this co
ompany Mobiilistic MIBS, tthis is Mobilistic Innovativ
ve Business Solutions, based
b out of
India,
00:11:06 Every yea ar, around this time, you kknow, in the Summer, the
e monsoonss come in Ind dia, and they
bring a lott of disease malaria, tyyphoid, chole
era and th
here is alwayys a shortage
e of relief
workers,
00:11:39 Here is Fa
an Appz, which looks at oof course the
e fan experie
ence, and... m
more than 80
00
companie es already on
n their way to
o a thousand of them in 57
5 different ccountries,
00:11:49 building th
heir applications on HANA A. It is increddibly inspirattional. So whhen you think
k about this
amazing, this inspirational work th ese sort of companies
c arre doing
29
00:13:51 the best iss in front of us,
u and our pprimary limita
ation is our im
magination. A
And the good
d news with
that is tha
at this is something that iss under our control.
c We can
c fix that.
30
www.sap.c
w com
2014
4 SAP AG or an SAP affiiliate company. All rights reserved.
No part of this publication may bbe reproduced or transmitted in any form or for
any purpose without the expresss permission of SAP AG or an SAP affiliate
company.
SAP an nd other SAP products annd services mentioned he erein as well as their
respective logos are trademarkss or registered trademarks of SAP AG (or an
many and other countries. Please see
SAP afffiliate company) in Germa
www.sap.com/corporate-een/legal/copyright/index.epx#trademark for
http://w
additional trademark informationn and notices. Some softwware products
marketted by SAP AG and its disstributors contain propriettary software
components of other software vevendors.
Nationa may vary.
al product specifications m
These materials are provided by SAP AG or an SAP P affiliate company for
informa ational purposes only, w without representation or warranty of any kind,
and SAP AG or its affiliated companies shall not be b liable for errors or
omissio ons with respect to the m materials. The only warrranties for SAP AG or
SAP afffiliate company productss and services are those that are set forth in the
express warranty statements aaccompanying such pro oducts and services, if
any. No othing herein should be cconstrued as constituting an additional warranty.
iated companies have no obligation to pursue
In particular, SAP AG or its affilia
any course of business outlinedd in this document or any related presentation,
or to deevelop or release any funnctionality mentioned therein. This document, or
any related presentation, and SA AP AGs or its affiliated companies strategy
and po ossible future developmennts, products, and/or platfform directions and
functionality are all subject to chhange and may be changed by SAP AG or its
ed companies at any timee for any reason without notice. The information
affiliate
in this document
d is not a commiitment, promise, or legal obligation
o to deliver
any ma aterial, code, or functional
ality. All forward-looking sttatements are subject
to varioous risks and uncertaintiees that could cause actual results to differ
materia ally from expectations. Reeaders are cautioned not to place undue
reliance on these forward-lookinng statements, which speak only as of their
dates, and they should not be reelied upon in making purc chasing decisions.