Professional Documents
Culture Documents
inside
1 Executive summary p5
2 Introduction p12
3 Circuit-switching and packet-switching
technologies p15
4 Challenges facing operators for their networks p29
5 Case study: comparison of costs and revenues
in OTN and MPLS networks p37
6 Conclusions p55
www.analysysmason.com 1
White paper audience • an explanation of multi-protocol label switching
(MPLS) and optical transport network (OTN)
This white paper has been written and designed for
technology and standards
operators’ executives to be able to take an informed
decision regarding their technical strategy for their • a business-case comparison for different network
core network. architectures to enable operators’ executives to
take an informed decision regarding their
White paper objectives
technical strategy for their core networks
The objectives of this white paper are to provide
• an insight into how their technical strategy for
operators’ executives with:
their core networks impacts the flexibility of
• a simple analogy with the highway and railway the business model they can adopt to remain
network to illustrate the concepts of utilisation of profitable.
resources, routing flexibility, reliability and impact
of traffic change in circuit-switched and packet-
switched networks, to better understand the key
issues associated with the exponential growth of
packet-switched traffic and the rapid decline of
legacy circuit-switched traffic in their networks
Copyright © 2012. The information contained herein is the property of Analysys Mason Limited and is provided
on condition that it will not be reproduced, copied, lent or disclosed, directly or indirectly, nor used for any
purpose other than that for which it was specifically furnished.
2
Contents
1 Executive summary 5
1.1 Key issues in the telecoms industry 6
1.2 Technology scenarios 8
1.3 Conclusions 11
2 Introduction 12
3 Circuit-switching and packet-switching technologies 15
3.1 Transport analogy 16
3.1.1 Railway analogy for circuit-switched networks 16
3.1.2 Highway analogy for packet-switched networks 18
3.2 Circuit-switched and packet-switched networks 21
3.3 OTN and MPLS 23
3.3.1 OTN 23
3.3.2 MPLS 24
3.4 Carrier-grade Ethernet services 26
4 Challenges facing operators for their networks 29
4.1 Bridging the gap between stagnating revenues and increasing network costs 30
4.2 Coping with the increasing unpredictability of traffic patterns 31
4.3 Maximising the use of network resources to optimise capex and opex 33
4.4 Maximising revenue opportunities 34
4.5 Guaranteeing appropriate levels of QoS for packet-switched traffic 35
5 Case study: comparison of costs and revenues in OTN and MPLS networks 37
5.1 Variables and assumptions 38
5.1.1 Network topology assumptions 39
5.1.2 Traffic matrix and traffic characteristics assumptions 39
5.1.3 Technology scenario assumptions 40
5.1.4 Scenario costing assumptions 42
5.1.5 Service pricing assumptions 43
5.2 Capex and revenue results 44
5.2.1 Provisioned capacity 45
5.2.2 Capex 45
5.2.3 Revenue 46
5.2.4 Capex efficiency (revenue per invested capex) 46
5.2.5 What does the result really mean in terms of capex and revenue? 47
5.3 Opex considerations 48
5.3.1 Scale of opex involved 48
5.3.2 Maintenance opex 49
5.3.3 Capacity planning and management opex 51
5.3.4 Power consumption opex 52
5.3.5 What does this result mean in terms of opex? 53
6 Conclusions 55
Annex A Traffic matrix 56
Annex B Detailed results 60
Annex C Glossary of terms 64
3
4
1 Executive summary
It shows that the choice of technology dramatically influences the business model that operators
can adopt for their core networks to remain profitable, and provides a business-case comparison
for different network architectures to enable operators’ executives to take an informed decision
regarding the optimum technical strategy for their core networks.
This white paper was developed in collaboration with a number of operators and equipment
vendors. All the assumptions used within this white paper were validated with them over the
period between May 2011 and October 2011.
5
WHITE PAPER
Optical transport network (OTN) and/or multi-protocol label switching (MPLS)? That is the question
“One of the main Today, it is widely accepted that the volume of packet-switched traffic is increasing
challenges for operators at an exponential rate in operators’ networks, and that in the next 10 to 20 years the
today is the increasing vast majority of traffic will be packet-based.
unpredictability of
traffic in terms of the This fast-changing paradigm in the telecoms Coping with the increasing unpredictability of traffic
industry has meant that operators must be more patterns
amount being carried flexible than ever to adapt their business model to
One of the main challenges for operators today is the
on their networks, address the following issues:
increasing unpredictability of traffic in terms of the
where it is coming • How to bridge the gap between stagnating amount being carried on their networks, where it is
from, and where it revenues and increasing network costs? coming from, and where it needs to go. The causes
for this unpredictability are multiple:
needs to go.” • How to cope with the increasing unpredictability of
traffic patterns? • The consolidation of data centres and the advent of
cloud computing allow content and service
• How to optimise the use of network resources to
providers to ‘migrate’ content and computing
minimise capital and operational expenditure
resources from one location to another, based on
(capex and opex)?
where they need to be consumed. This creates
• How to maximise revenue opportunities? substantial shifts in traffic patterns as sources and
sinks of information can change instantaneously.
• How to guarantee appropriate levels of quality of
service (QoS) for various kinds of packet-switched • The increased mobility of content users presents
traffic (in particular, guaranteeing high-quality additional challenges. Until recently, there was a
voice services)? clear relationship between the user and the user’s
location when accessing the network, as everybody
Bridging the gap between stagnating revenues and was physically ‘tethered’ to the network. Today’s
increasing network costs radio access networks are increasingly capable of
Many operators are experiencing eroding profit supporting high-bandwidth applications, including
margins because their costs and revenues are streaming video, and a plethora of mobile devices
increasing at different rates. Revenues are allow people to consume content no matter where
stagnating, mainly due to a decrease in voice they are. As a result, consumers have detached
revenues and increasing competition in other, if not themselves from the network; they are mobile and
all, services. Costs, on the other hand, have they can do things on the move that they used to
continued to rise as a result of the infrastructure only be able to do sitting in front of their ‘attached’
expansion required to support new data services. computers.
Most importantly, the traditional service provider • Exceptional events such as sports events (e.g.
business model is being challenged, primarily football finals, Olympics, etc.) and other events
because of the de-integration of the traditional such as political elections, can generate large
vertical model, forcing network service providers to short term demands between particular network
carry an ever-increasing amount of traffic over the nodes (i.e. telecommunication nodes serving the
top (OTT). OTT services are services that are created different venues of that event).
outside of the operator’s network. Yet, service
providers have to carry this traffic without extracting The net effect of mobility and cloud computing is that
any additional income, as all the service revenue aggregation networks become less efficient.
goes to the owner of the content. Service provider Aggregation networks are static and are built based
revenues are restricted to the network access, which on knowing where the users are, where the content
is sold as flat-rate bandwidth plans. As a result, the is stored, and where the applications are running. All
core of the network has become a cost-centre of this is now fluid and dynamic, and hence the core
commodity, and the goal becomes to relentlessly transport network needs to provide flexible, ad hoc
pursue a strategy and architecture that takes every aggregation, and packet-switched networks provide
single bit of cost out of that network, yet making sure the optimum technology to achieve this. In marked
that it remains flexible enough to handle all the contrast, circuit-based technology is more suitable
varying traffic it has to carry, while meeting or for static traffic and is not well adapted for ever-
exceeding the service level agreements (SLAs) in changing traffic patterns.
place with customers.
6
“Operators are seeing rapid growth in demand for carrier Ethernet services, for both business Nowadays, packet networks
and wholesale services. Industry analysts, including Infonetics Research and Ovum, continue routinely carry voice traffic,
as illustrated by the adoption
to forecast strong growth in worldwide carrier Ethernet services – the market is currently of next generation networks
(NGNs) by operators
worth USD20 billion and is set to grow to USD50 billion by 2014.1” throughout the world.
Optimising the use of network resources to carrier Ethernet services – the market is currently
minimise capex and opex worth USD20 billion and is set to grow to USD50
billion by 2014.1
Operating in such a difficult landscape requires more
dynamic intelligence in the network and optical The most dramatic growth in carrier Ethernet
layers. More immediately, maximising the use of services is coming from mobile backhaul. Infonetics
existing assets is of paramount importance for Research forecasts that Ethernet microwave
profitability. However, circuit-switched infrastructure revenues will grow at a compound annual growth
is inefficient at carrying packet-based traffic. rate (CAGR) of 41% over the period 2011–2015. The
A typical packet-based traffic flow will only peak at problem facing mobile carriers – on top of downward
its maximum bandwidth in short, infrequent price pressures – has been the surge in mobile data
intervals, and most of the time, will have a traffic since the iPhone was launched in 2007, plus
throughput significantly lower than the maximum the fact that the smaller footprint of 3G cell sites
bandwidth. Therefore, dedicating for example 1Gbit/s requires more cell sites with scalable backhaul
of capacity at all times for a GbE service traffic flow capabilities. Another key driver for carrier-grade
would result in a significant under-utilisation of Ethernet services has been video applications. For
network resources. Operators using circuit-switched example, Netflix is now dominating North America
technology (such as optical transport network or bandwidth demand and smartphones are pushing up
OTN) need to allocate fixed-capacity circuits to the use of mobile video.
transport packet-switched traffic, which results in
Operators are rapidly responding to this increase in
‘stranded’ capacity on those circuits that no other
demand by deploying packet-based infrastructure in
services can use.
their networks to support the delivery of carrier-
In marked contrast, the ability of packet-switched grade Ethernet metro services. In marked contrast,
networks to aggregate traffic and use a pool of circuit-switched technology can only provide a subset
shared capacity means that trunk links on packet- of carrier Ethernet services, therefore reducing the
switched networks typically require much less opportunity for revenue.
capacity than would be needed from an equivalent
Guaranteeing appropriate levels of QoS for various
circuit-switched network. This effect is known as
kinds of packet-switched traffic
statistical multiplexing gain, where some traffic
flows will peak and others will trough, compensating Packet-switched networks have evolved dramatically
for one another, and therefore requiring much less over time. In the early days of packet switching, all
capacity overall than if the networks were networks were ‘best effort’, which meant that
dimensioned for the peak traffic requirements. different types of traffic were all carried with the
same priority and were all subject to the same
Put another way, with a circuit-switched network, an
degradation in performance when a network
operator can only sell the provisioned bandwidth only
congestion occurred. Consequently, there was a
once, while packet-based capacity can be sold
justified scepticism as to whether packet networks
multiple times limited only by the amount of
were good enough to carry voice. Nowadays, packet
statistical gain that can be achieved.
networks routinely carry voice traffic, as illustrated
Maximising revenue opportunities by the adoption of next generation networks (NGNs)
by operators throughout the world. The focus has
Operators are seeing rapid growth in demand for
shifted entirely away from whether packet networks
carrier Ethernet services, for both business and
are capable of carrying high-priority traffic
wholesale services. . The challenging economic
demanding high QoS; instead, the main interest is in
climate that currently exists is further driving the
using the technology to carry both high-priority and
need for intelligent and efficient networks. Industry
low-priority traffic over the same network at the
analysts, including Infonetics Research and Ovum,
lowest possible cost.
continue to forecast strong growth in worldwide
1 Total Telecom (August 2011), Carrier Ethernet key to telecoms growth. Available at http://www.totaltele.com/view.aspx?ID=467030. 7
WHITE PAPER
Optical transport network (OTN) and/or multi-protocol label switching (MPLS)? That is the question
“In order to estimate In response to these challenges, operators can implement different technical
the revenue associated strategies based on different core network architectures. In this white paper, we
with each network consider three possible strategies to handle growth in packet-switched traffic and
architecture, we maintain service quality, namely the following:
1 Service 1
2 3 4
1 2 3 4 Service 1
switching (MPLS) technology, where all switching deliberately do not provide any absolute 1 2capex3 4or Service 5
Ethernet services when in the core network occurs on a packet basis at opex, nor any absolute1 revenues
2 3due to the 4 Service 6
17 Gbit/s 17 Gbit/s
8 b)
2 Relative to scenario 2- OTN network.
Core Network
“We have seen that an MPLS operator does not need to provision as much additional However, the business case
for choosing the most
capacity to cope with increasing demand for bandwidth. This leads to a much more efficient architecture is
sustainable business model, where capital costs are dissociated to a greater extent from dictated by profitability. We
found that, for every dollar of
bandwidth demand.” capex invested in the
network, an MPLS network
will provide 58% additional
revenues for the operator
compared to an OTN
100%
Network for a P/A ratio of 2,
and 81% for a P/A ratio of 3.
relative capex compared to an OTN network
90%
80%
45%
70%
18%
60%
100% 100%
100%
network provisionned capacity compare to OTN network
100% 50%
relative capex compared to an OTN network
90%
90%
relative capex compared to an OTN network
50% 50% 0%
90%
relative capex compared to an OTN network
50% 0% 50%
50% 90%
0% MPLS (scenario
0% 1) OTN (scenario 2) MPLS + OTN
40% 40%
35% MPLS (scenario 1) 35%
OTN (scenario 2) MPLS + OTN 1) OTN(scenario
MPLS (scenario 18% (scenario3)2) MPLS + OTN 1.3
40% 80%
35% 35% (scenario 3) 30% (scenario 3)
30% 56% 54%
Total DWDM port cost Total MPLS or OTN port cost
30% 70% 20%
20% Total DWDM port cost Total MPLS or OTN port cost
29%
20% 60% 10%
10%
10% 50% 0%
0%
MPLS (scenario 1) OTN (scenario 2) MPLS + OTN1.3
MPLS (scenario 1)40% OTN (scenario 2) MPLS + OTN (scenario 3)
0% 35% 100% 35%
(scenario 3)
1.3
an OTNprovisionned
MPLS 100%
(scenario 1) OTN (scenario 2) MPLS + OTN
30% (scenario 3)
network
70%
0%
60% 100% (scenario 1)
MPLS OTN (scenario 2) MPLS + OTN 1.3
100%
(scenario 3)
100% 180%
Relative revenue compared to an OTN network
50%
90% 100% 158%
relative capex efficiency compared to the OTN relative capex efficiency compared to the OTN
160%
Relative revenue compared to an OTN network
40%
80% 90%
75% 75%
30% 140%
70% 80% 75% 75%
20% 120%
60% 70% 100% 105%
100% 100%
10% 180% 100%
50%
network
60%
158%
relative capex efficiency compared to the OTN
90%
0% 160%
40% 180% 80%
50% MPLS (scenario 1) OTN (scenario 2) MPLS + OTN
80% 75% (scenario
75% 3) 158%
relative capex efficiency compared to the OTN
40%
0% 0%
20% 1) 80%
MPLS (scenario 1) OTN (scenario 2) MPLS + OTN MPLS (scenario OTN (scenario 2) MPLS + OTN
(scenario 3) (scenario 3)
60%
0%
MPLS (scenario 1) OTN (scenario 2) MPLS + OTN
40% (scenario 3)
20%
0%
MPLS (scenario 1) OTN (scenario 2) MPLS + OTN
(scenario 3)
Figure 1.4: Revenue for different scenarios for P/A=2 Figure 1.5: Main result of case study: revenue per
[Source: Analysys Mason] invested unit capex for different scenarios for P/A=2
[Source: Analysys Mason]
9
WHITE PAPER
Optical transport network (OTN) and/or multi-protocol label switching (MPLS)? That is the question
“ An MPLS network capacity required in an MPLS network means that it compared to an OTN operator. In this context, it is
has significantly fewer ports than an OTN (for the clear that MPLS technology will be more suited to
consumes 40% less same traffic matrix). Overall, the capex for an MPLS the unpredictable nature of future traffic.
energy compared to an network is only 47% of the capex for an OTN for a P/A
We have seen that an MPLS operator does not need
OTN, which will help of 2, and decreases to 42% for a P/A of 3.
to provision as much additional capacity to cope with
an MPLS operator In order to estimate the revenue associated with increasing demand for bandwidth. This leads to a
each network architecture, we conducted market much more sustainable business model, where
reduce its energy bill research on what operators charge for Ethernet capital costs are dissociated to a greater extent from
and meet its green services when provided on an MPLS versus OTNs. bandwidth demand. This can be achieved by either:
objectives.” We found that MPLS-based operators typically
• implementing different classes of service for
charge 25% less for a 1:3 contended service than
revenue-generating compared to non-revenue-
OTN-based operators (as illustrated in Figure 1.4).
generating traffic such as OTT applications and
The main difference between the two services is that
peer-to-peer (e.g. best-efforts only for the
the MPLS operator will typically provision capacity
non-revenue-generating traffic)
(and therefore guarantee) for just one third of the
• increasing the contention ratio in the network, and
peak bandwidth, whereas the OTN operator will
thereby offering more and more services with the
provision capacity for the full peak bandwidth.
same provisioned capacity, but with a reduced SLA.
However, the business case for choosing the most
Opex results
efficient architecture is dictated by profitability. We found
180% that, for every dollar of capex invested in the network, an
180% In our modelling we have considered the following
160% MPLS
160% network will provide 58% additional revenues for
opex elements: maintenance, capacity planning and
140% the operator compared to an OTN for a P/A of 2 as
140% 200% 200%
energy consumption. Figure 1.6 and Figure 1.7
180%
120% illustrated
120% in Figure 1.5, and 81% when the P/A ratio 180%
represent the relative maintenance opex for the
160% 160%
100% 100%
increases to 3. different scenarios with P/A ratios
140% 140% of 2 and 3. In Figure
80% 80%
158% 158% Capex efficiency
120% 1.6 and Figure 1.7, we differentiate 120% between Level 1
45% 45%efficiency
Capex
60% 18% It is60%
interesting
105% to Revenue
note that18% in a real network105% it will Revenue
100% 100% 100%100% 100% 100% 100% and Level 2 maintenance which 100%is usually performed
40% not be
40% possible to forecast the P/A ratio with any 181% 181%
18% 75% 75% 18%
Total DWDM port
56%
75%cost 75% Total
80%DWDM port cost
by ‘in-house’ operational
45% teams 80%from the moreCapex 45%
efficiency
20% 37% 37% 29%
56%
accuracy,
20% 37%mainly 37% due to the
54%orincreasingly
Total MPLS OTN port cost Total
60%MPLS or OTN port cost 60% 117% Revenue 16% 1
54% 29% complex
100% Level 3 and Level
16% 4 maintenance
100%100% activities 100% 100%
0% unpredictable
0% nature of traffic demand. The majorityCapacity
Capacity 40%
100%
40% Total DWDM port75%cost 75%
which are usually16% performed by the equipment vendor
75% 75% 16%
56% 49%
MPLS
MPLS
MPLS
MPLS
OTN
OTN
OTN
OTN
MPLS
MPLS
MPLS
OTN
OTN
OTN
OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
20% 35% 35% 26% 20% 35% Total MPLS or OTN port cost
35% 26%
in the form of a contract with the service provider.
increase over time, even for the highly aggregated 0% 0% Capacity
MPLS
MPLS
MPLS
MPLS
OTN
OTN
OTN
OTN
MPLS
MPLS
MPLS
MPLS
OTN
OTN
OTN
OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
trunk links in the core network, but no operator is
MPLS + OTN
MPLS + OTN
MPLS + OTN
Figure 1.8 and Figure 1.9 show the relative energy MPLS + OTN
able to predict by how much. In this dynamic consumption.
Provisioned Total capex Revenue Provisioned
Revenue per Total capex Revenue Revenue per
capacity scenario, MPLS
capacity
unit capex networks provide an even better
unit capex
prospect than OTN because the higher the P/A ratio, From the above analysis, it is clear that an MPLS
Provisionedwill Total capex significantly
Revenue Provisioned
Revenue per Total capex Revenue Revenue
the more profitable an MPLS operator will be network
capacity
require less opex than
capacity
unit capex
an unit cape
OTN for the following reasons:
5.9 5.9
0% 0% 0% 0%
MPLS OTN MPLS + OTN MPLS OTN MPLS + OTN MPLS OTN MPLS + OTN MPLS OTN M
Level 1 and 2 maintenance opex Level 1 and 2 maintenance opex Level 1 and 2 maintenance opex Level 1 and 2 maintenance o
Level 3 and 4 maintenance opex Level 3 and 4 maintenance opex Level 3 and 4 maintenance opex Level 3 and 4 maintenance o
10 Figure 1.6: Maintenance opex comparison for P/A =2 Figure 1.7: Maintenance opex comparison for P/A =3
[Source: Analysys Mason] [Source: Analysys Mason]
5.10 5.10
50%
Relative power consumption
% 0% 0% 0%
MPLS OTN MPLS
MPLS + OTN OTN MPLS + OTN MPLS OTN MPLS
MPLS + OTN OTN MPLS + OTN
Figure 1.8: Power consumption for a P/A of 2 Figure 1.9: Power consumption for a P/A of 3
[Source: Analysys Mason] [Source: Analysys Mason]
1.3 Conclusions
200% 200%
180% 180%
investment
54% MPLS or OTN port cost
Total for60%oranOTNoperator,
Total MPLS port cost
100%
for the
16% following
60%
100% 100%
100%
reasons:
117% Revenue
16%
100% 100%
117% Revenue
Total DWDM port cost
Capacity Capacity 40% 16% 75% 40% 75% 16% Total DWDM
75% port cost
75%
56% 49%
MPLS
MPLS
OTN
OTN
OTN
56% 49%
MPLS + OTN
MPLS + OTN
MPLS + OTN
20% 35% 35% 26% 20% 35% 35% 26% Total MPLS or OTN port cost Total MPLS or OTN port cost
• MPLS enables an operator to dissociate traffic
0% 0% • An MPLS-based operator will incur less opex
Capacity Capacity
MPLS
MPLS
MPLS
MPLS
OTN
OTN
OTN
OTN
MPLS
MPLS
MPLS
MPLS
OTN
OTN
OTN
OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
bandwidth and capacity provisioning, which is key compared to an OTN-based operator, which will
MPLS + OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
in controlling costs and therefore breaking away have significantly more ports to operate and
capex Revenue Revenue per
from eroding margins.
unit capex maintain.
• MPLS makes it possible to differentiate between
Provisioned Total capex • MPLS architecture and technology enable an
Revenue Provisioned
Revenue perTotal capex Revenue Revenue per
different traffic types (e.g. revenue-generating
capacity capacity operator to significantlyunit
unit capex capex its power
reduce
x efficiency
nue versus non-revenue-generating) and to adjust consumption, saving costs and helping it to meet
DWDM port cost their capex spending accordingly. Capex efficiency its Capex
greenefficiency
agenda.
Revenue Revenue
MPLS or OTN port cost • MPLS provides the ability for the operator to offer Therefore, a native MPLS architecture is the only one
Total DWDM port cost Total DWDM port cost
city the full range of carrier Ethernet services, which whichTotal
offers
Total MPLS or OTN port cost MPLSthe flexibility
or OTN port cost for operators to adapt their
represent one of the fastest-growing markets. business model in line with the changes currently
Capacity
Capacity
• MPLS is more suited to accommodating changes being experienced in the telecoms industry, by
in traffic patterns, which, with the advent of cloud providing a way to differentiate revenue-generating
computing and the increase in user mobility, is and non-revenue-generating traffic, and to control
becoming an increasing issue. capex and opex.
• MPLS will enable the operator to generate
between 58% and 81% more revenues than OTN, 11
for the same capex investment.
WHITE PAPER
Optical transport network (OTN) and/or multi-protocol label switching (MPLS)? That is the question
2 introduction
“For most operators,
the following question
remains: how can
they implement a
network architecture
that provides
enough flexibility to
accommodate the
expected growth in
packet-switched traffic, Today, it is widely accepted that the volume of packet-switched traffic is increasing
while addressing at an exponential rate in operators’ networks, and that in the next 10 to 20 years the
quality-of-service vast majority of traffic will be packet-based. The main factors driving the increase
(QoS) issues by type in packet-switched traffic are:
of traffic and at the
same time minimising • market growth in both fixed and mobile broadband This white paper explores the different technical
costs to increase their services options that are available to operators to cope with
the explosion of packet-switched traffic on their core
profitability?” • availability and attractiveness of affordable smart
network. There are three strategies available to
phones and tablets (e.g. iPhone and iPad)
operators to enable them to handle the expected
• growth in video applications, especially over-the growth in packet-switched traffic on their core
top (OTT) applications such as Netflix, BitTorrent, networks whilst maintaining service quality:
YouTube, and catch-up TV applications such as the
• implement a circuit-switched network based on
BBC iPlayer
OTN technology, whereby switching in the core
• growth in the business segment network occurs on a circuit basis and where the
switching of packets only occurs at the edge of the
• migration of legacy circuit-switched services to network
packet-switched services (e.g. mobile backhaul,
enterprise services and the public-switched • implement a packet-switched network based on
telephone network or PSTN). MPLS technology, whereby switching in the core
network occurs on a packet basis
Yet, for most operators, the following question
remains: how can they implement a network • implement a mix of OTN and MPLS technology,
architecture that provides enough flexibility to whereby some core nodes can switch both packets
accommodate the expected growth in packet- and circuits and some other nodes switch either
switched traffic, while addressing quality-of-service packets or circuits.
(QoS) issues by type of traffic and at the same time
We assess the viability of the business case for each
minimising costs to increase their profitability? The
of these three strategies by comparing their costs
answer to this question is not straightforward: it
and associated revenues.
depends, to a large extent, on the type of operator,
the services provided and the legacy network of the
said operator.
12
“This white paper explores the different technical options that are available to operators to cope
We assess the viability of
with the explosion of packet-switched traffic on their core network. There are three strategies the business case for each
available to operators to enable them to handle the expected growth in packet-switched traffic of these three strategies
by comparing their costs
on their core networks whilst maintaining service quality..” and associated revenues.
Structure of this white paper • Section 5 assesses the business case for the
different technology strategy options available to
The remainder of this white paper is laid out as
operators by considering three reference
follows:
scenarios:
• Section 3 introduces the concepts of circuit
— scenario 1: native MPLS network underpinned by a
switched and packet-switched networks,
dense wavelength division multiplexing (DWDM)
illustrating them by an analogy with the highway
transport layer
and railway network. This analogy is then used to
explain how circuit- and packet-switched networks — scenario 2: native OTN underpinned by a DWDM
(such as MPLS networks) operate. transport layer
• Section 4 covers the major challenges currently — scenario 3: MPLS network with OTN bypass and
facing operators for their core networks and how underpinned by a DWDM transport layer.
these challenges may impact their technical
• Section 6 presents our conclusions from the study.
strategy decisions in order to meet their business
objectives. These challenges include: This white paper also includes the following annexes
containing supplementary documentation:
— How to bridge the gap between stagnating
revenues and increasing network costs? • Annex A includes the characteristics of all traffic
flows considered for the different technology
— How to cope with the increasing unpredictability of
scenarios studied in this paper
traffic patterns?
• Annex B provides the detailed results of our
— How to maximise revenue opportunities?
assessment of the business case of the different
— How to optimise the use of network resources to technology scenarios
minimise capital and operational expenditure
• Annex C includes a glossary of the terms used
(capex and opex)?
throughout this white paper.
— How to guarantee QoS for packet-switched traffic
(in particular, voice)?
13
14
3 Circuit-switching
and packet-switching
technologies
This section introduces the concepts of circuit-
switching and packet-switching. It is structured
as follows:
• Section 3.1 illustrates these concepts by an analogy with the highway and railway network
• Section 3.2 explains the concept of circuit switching and packet switching technology in more detail in
the context of a telecommunications network
• Section 3.3 provides an introduction to Optical Transport Network (OTN) and Multi-Protocol Label
Switching (MPLS) technology
15
IP/Ethernet
1 2 3 4 Service 1 LSP-a OTU-y
IP/Ethernet
1 2 3 4 LSP-b ODU
Service 2
IP/Ethernet mapper
ODU OTU-x λ
1 2 LSP-c
34 Service 3
OTU-v
1 2 3 4 Service 4
OTH interfaces OTU-z
1 2 3 4
WHITE PAPER Service 5
Client λ interfaces
Optical1 transport
2 network
3 (OTN) and/or
4 multi-protocol
Service 6 label switching (MPLS)? That is the question
1 2 3 4 Service 7
1 2 3 4 Service 8
IP/Ethernet
IP/Ethernet
LSP-a
LSP-b ODU
mapper
ODU
OTU-y
OTU-x λ
LSP-c
1 1 1 1 21 21 2 32 13 232 3 3 2 1 34 2 4 433 4 4 4 4 4 Shared bandwidth OTU-v
Barcelona Madrid
Carriage A Carriage B Carriage C
Keys
17 Gbit/s
Locomotive
Paris-Barcelona average traffic
Locomotive
Figure 3.1: Example of train utilisation during peak and non-peak time hours
[Source: Analysys Mason, 2011]
16
Core Network
IP/Ethernet mapper Layer
LSP-c
Service 3
OTU-v
Service 4 WDM
OTH interfaces OTU-z
Layer
Service 5
Service 8
IP/Ethernet
LSP-a London Paris Barcelona Madrid
OTU-y
IP/Ethernet
LSP-b ODU Switching
IP/Ethernet mapper
ODU OTU-x λ Layer
LSP-c
Shared bandwidth OTU-v
WDM
OTH interfaces OTU-z
Layer
WDM Bypass
Client λ interfaces
IP/Ethernet
LSP-a OTU-y
IP/Ethernet
LSP-b ODU
IP/Ethernet mapper
ODU OTU-x λ
LSP-c
OTU-v
“Typically in a railway network, the train operator
OTH interfaces OTU-z defines a fixed set of routes to operate, where In circuit-switched networks,
each route is defined by the city of origination, the city of destination and all the intermediate circuits are equivalent to R8
Client λ interfaces train routes, which are
cities along that route. Therefore, if the operator has not defined a service between two end established between source
and destination nodes, via a
points, then it is not possible to travel on that route because no trains are available.” number of transit nodes.
R2
20Mbps
40
60 27
R1 49
Station A Station B Station C
Train 1
Train 2 Keys
RS
Train service RS
Passenger demand 25 Ba
Nobody can travel from
Rail infrastructure 25 Lab
Station A to Station D
because no train service is
Station D
available
Train 3
As illustrated in Figure 3.2, no passenger will be able networks consists of grouping circuits together to 30
to travel from Station A to Station D (assuming that increase the utilisation of the trunk link. R2
“Therefore, circuit- We discuss each of these in turn. dramatic increase in traffic, but this would incur
significant capex (i.e. the cost of additional
switched networks > Changes in the demand for a particular route
carriages). This issue is explored in more detail in
have limited flexibility If, for a particular route, there is a change in the Section 4.4.
to cope with significant demand due to, for example, an exceptional sports
> Changes in the route
event such as a football final, additional capacity in
changes in the the form of more passenger carriages will be If we consider our example in Figure 3.2, and that for
volume of traffic required to accommodate the exceptional demand some reason there was a surge in demand to go
for that service, as many more passengers will want from Station A to Station D, then both Train 1 and
on these networks.
to use that service on that day. If this exceptional Train 3 would need additional carriages to cope with
Accommodating these demand can be predicted in advance, then the train the excess in demand. One solution to prevent
changes in real time operator will be able to add carriages to the service, passengers from changing trains would be to group
but this will involve some opex (i.e. the cost of them in dedicated carriages in Station A and then
can only be achieved attaching new carriages to the train, which is very detach these carriages from Train 1 and attach them
if the circuits were labour-intensive). However, if the exceptional to Train 3. Detaching and re-attaching a carriage to a
demand cannot be predicted in advance, unless the different train is very labour-intensive and incurs
over-provisioned in the
number of carriages was over-provisioned in the first significant opex.
first place, incurring place, the train operator will not be able to
The same is true in circuit-switched networks, where
significant capex.” accommodate all the passengers wanting to travel
traffic destined to a particular node can be grouped
as there will not be enough time to plan the addition
together on larger circuits and where circuits can be
of carriages to the existing route.
switched so that it reaches its destination.
The situation for circuit-switched networks is similar.
Therefore, circuit-switched networks have limited
Circuits are provisioned for a given peak capacity
flexibility to cope with significant changes in the
and, if the demand were to suddenly increase, larger
volume of traffic on these networks. Accommodating
circuits would need to be provisioned, involving some
these changes in real time can only be achieved if the
re-configuration and associated opex. Alternatively,
circuits were over-provisioned in the first place,
the operator could over-provision the circuit-
incurring significant capex.
switched network in the first place to cater for any
R8 R3 R4
10
30 30
R2 R5
d
Carriage A Carriage B Carriage C
Keys 12 25 8
40 50
25
Car Lorry
Locomotive 60 8
80
R1 12
41 41
Round about R6 R7
a) 95%
Tributary utilisation during peak time
road Traffic
Carriage A Carriage C
Car Lane 2 flow:
Carriage B 20Mbit/s
Car Lane 1
Locomotive 17.5Gbit/s 18.5Gbit/s 25 16.5Gbit/s
Bandwidth (Mbit/s) available on each link
Trunk Road
Taxi and bus Lane 25 Label value assigned to the traffic flow
16.5Gbit/s Label Switched Path (LSP)
b) 40% utilisation during non-peak time
Emergency lane
16.5Gbit/s
Keys
Core Network London Paris Barcelona 17Gbit/s Madri
Car Lorry
17Gbit/s
Stranded capacity
NEW FIG. 5.2
OTN Circuit
19
“MPLS allows traffic The same idea applies to packet-switched networks. Impact of traffic change
If no QoS (or priority) is applied to the traffic, then the
engineering to be traffic from tributary links at a given network node
Similarly to what we did in the case of circuit-
implemented in packet- switched networks, we consider two types of
will all be served on a ‘first come, first serve’ basis,
changes in the case of packet-switched networks:
switched networks, meaning that it is extremely difficult to predict how
long it will take for a packet to get to its destination • changes in the demand for a particular service
thus giving operators as it is highly dependent on the overall volume of
• changes in the route.
the ability to guarantee traffic. This is usually described as a ‘best effort’
policy. For time-sensitive traffic such as voice, this We discuss each of these in turn.
a minimum QoS for situation would clearly be unacceptable as it would
the traffic transported hamper the interactivity in the phone conversation. > Changes in the demand for a particular route
along each LSP.” Most modern packet-switched networks are no Using the road analogy, provided that the increase in
longer best effort, and can prioritise time-sensitive traffic on one of the tributary roads is compensated
traffic (such as voice and video) over non-time by a decrease in traffic on another tributary road, no
sensitive traffic (such as Internet browsing). The congestion will occur on the trunk road.
Internet protocol (IP) has long had the ability to However, an increase in traffic on a tributary road
define different classes of service for different traffic may cause that road to become congested. If priority
types. In our analogy, this is equivalent to having lanes are implemented, the emergency vehicles will
some prioritisation at the roundabout so that all not be affected by that congestion, provided that the
ambulances and police cars can jump the queue and priority lane has enough capacity to support all
do not incur unreasonable delays. emergency vehicles travelling on that lane.
Coming back to the highway analogy, creating This is similar to packet-switched networks, which
priority lanes without considering their respective exploit statistical multiplexing (not all traffic streams
capacities may not be enough to ensure the time of will peak at the same time). Also, as explained
arrival for the different traffic types. The solution to elsewhere in this white paper, MPLS enables
this problem is to ensure that for each traffic type the operators to define express paths so that different
lanes are dimensioned to accommodate their levels of QoS (such as bandwidth) can be guaranteed
respective traffic. In order to do this, one needs for different types of traffic.
visibility of the capacity on the road along every
segment, and one must ensure that the capacity on > Changes in the route
every lane is sufficient to accommodate all types of
As mentioned previously, vehicles can go wherever
traffic even during peak time hours.
they want, using any road they want, as long as the
In packet-switched networks, MPLS provides the driver follows the road signs for his/her destination
ability to define express paths (equivalent to lanes in along the road.
our analogy) between any two nodes in the network,
In the packet-switching paradigm, packets of data
and ensures that sufficient bandwidth is allocated to
(vehicles in our analogy) can be routed to any
each express path in the network to guarantee a
destination using the ‘routing protocol’, which is the
minimum QoS for the traffic carried on it.
equivalent to a driver following the road signs in
Establishing express paths (called label-switched
our analogy.
path or LSP) on the basis of a holistic view of the
network characteristics (e.g. available bandwidth, Therefore, packet-switching is more naturally
used bandwidth) is often referred to as traffic suited to cope with changes in the volume of traffic
engineering. MPLS allows traffic engineering to be due to its statistical multiplexing and to the fact
implemented in packet-switched networks, thus that dynamic routing is inherent to packet-switched
giving operators the ability to guarantee a minimum networks.
QoS for the traffic transported along each LSP.
The combination of defining classes of service and
LSPs is extremely powerful as, in effect, it combines
the advantages of packet-switched and circuit-
switching technologies; the traffic from different
services still uses the same shared resources, and
each service can be guaranteed a minimum QoS in
terms of delay and throughput, for example. This
unique proposition explains why most operators have
implemented some form of MPLS in their packet-
switched networks.
20
3.2 Circuit-switched and packet-switched
networks
Legacy PSTNs are based on circuit-switching technology, which allocates a IP traffic is expected to show
dedicated physical path to each voice call and reserves an associated amount of the opposite trend, growing
from 30–50% of all traffic
dedicated bandwidth (usually a PSTN voice channel has a bandwidth of 64kbit/s) carried on the core transport
networks today to more than
across the network. 90% by 2016.
This bandwidth is dedicated to the call connection for route or the time it took for the packet to arrive at its
the duration of the call whether or not any audio destination. This new concept meant that information
voice is being exchanged between the callers. could now be sent in small packets through a shared
network. The main advantage of this technology was
Consequently, network planners and designers have
that, whenever no information needed to be sent, no
to dimension their circuit-switched networks
resources were utilised, enabling other traffic
according to the number of calls in the busy hour,
streams to use these resources instead. This is
factoring in a blocking probability in their design to
illustrated in Figure 3.6.
keep costs down. The blocking probability represents
the probability that a caller will not be able to make Today, packet-switching technology continues to be
his/her call because there are not enough circuits in the technology of choice for the Internet, and IP is at
the network to accommodate every single user to its centre.
make a phone call at the same time.
According to a recent study conducted by ACG
Because PSTN services were the dominant services research,2 the ratio of time-division multiplexing
on legacy networks, operators have built their trunk (TDM) to IP traffic will dramatically change over the
network (core networks) to link different towns using next five years. Today, TDM-encapsulated traffic still
circuit-switching technology. This is illustrated below represents 50–70% of all traffic carried on the core
in Figure 3.5. transport networks, and the ACG study indicates that
this ratio will decrease to 10% by 2016. IP traffic is
In the 1960s, the advent of the Internet created a
expected to show the opposite trend, growing from
disruptive communications technology known as
30–50% of all traffic carried on the core transport
packet switching. IP emerged from a military
networks today to more than 90% by 2016. It is
program (DARPAnet) which was developed to
interesting to note that, according to most operators
maximise the probability that packets (information or
interviewed for this white paper, the majority of
data) were guaranteed to arrive at their destination
traffic within TDM circuits is packet-based. The
irrespective of the state of the network (e.g. if one
findings of the ACG study are illustrated below in
route had been blown up), but did not guarantee the
Figure 3.7.
1 2 3 4 Service 1 IP/Ethernet
1 2 3 4 Service 1 1 2 3 4 IP/Ethernet
1 2 3 4
Service 1 LSP-a OTU-y
Service 1 IP/Ethernet LSP-a OTU-y
1 2 3 4 Service 2 IP/Ethernet
LSP-b
1 2 3 4 Service 2 1 2 3 4 Service 2 LSP-b ODUODU
1 2 3 4 Service 2 IP/Ethernet
ODU OTU-x
1 2 34 mapper ODU OTU-x
1 2 34 Service 3 IP/Ethernet
LSP-c mapper
Service 3 1 2 34 Service 3 LSP-c
1 2 34 Service 3 OTU-v
1 2 3 4 Service 4 OTU-v
1 2 3 4 Service 4 1 2 3 4 Service 4
1 2 3 4 Service 4 OTU-z
OTH interfaces OTU-z
OTH interfaces
1 2 3 4 Service 5
1 2 3 4 Service 5
1 2 3 4 Client λ interfaces
1 2 3 4
Service 6
Service 6 Client λ interfaces
1 2 3 4 Service 7
1 2 3 4 Service 7
1 2 3 4 Circuit 1 1 2 3 4 Service 8
1 2 3 4 Circuit 1 1 2 3 4 Service 8
IP/Ethernet
IP/Ethernet
1 2 3 4 Circuit 2 LSP-a OTU-y
1 2 3 4 Circuit 2 IP/Ethernet LSP-a OTU-y
IP/Ethernet
LSP-b
LSP-b ODUODU ODU
1 2 34 Circuit 3 IP/Ethernet mapper ODU OTU-x
1 2 34 Circuit 3 IP/Ethernet
LSP-c mapper OTU-x
LSP-c
1 2 3 4 Circuit 4 1 1 1 1 21 21 2 32 13 232 3 3 2 1 34 2 4 433 4 4 4 4 4 OTU-v
1 2 3 4 Circuit 4 1 1 1 1 21 21 2 32 13 232 3 3 2 1 34 2 4 433 4 4 4 4 4
Shared bandwidth
Shared bandwidth OTU-v
OTH interfaces OTU-z
OTH interfaces OTU-z
Circuit switched
Circuit network
switched network Packet switched network
Empty timeslot Packet switched network Client λ interfaces
Empty timeslot Client λ interfaces
Total capacity
Total = Sum
capacity of allofcircuits
= Sum capacity
all circuits capacity Total capacity
Total << Sum
capacity of allofservice
<< Sum peak
all service capacity
peak capacity
IP/Ethernet
IP/Ethernet
LSP-a OTU-y
IP/Ethernet LSP-a OTU-y
IP/Ethernet
LSP-b
LSP-b ODUODU ODU OTU-x
IP/Ethernet mapper ODU OTU-x
IP/Ethernet
LSP-c mapper
LSP-c
OTU-v
OTU-v
OTH interfaces OTU-z
OTH interfaces OTU-z
Figure 3.5: Circuit-switching
London
London
Paris
Paris concept
Barcelona illustration
Barcelona Madrid
Madrid Figure 3.6: Packet-switching example
MPLS Client λ interfaces
MPLS
[Source: Analysys Mason]
Switching
Switching [Source: Analysys Mason] Client λ interfaces
Layer
Layer
OTN
2 ACG OTN 21
Research
Switching
Switching
(June 2010), OTN Survey.
Layer
Layer
WDM
WDM
Layer
Layer
17Gbit/s
capacity
NEW FIG. 5.2
OTN Circuit
3.2 Circuit-switched and packet-switched networks continued
“There are a number With an increasing pressure to reduce costs, most contrast, when a user downloads an Internet page, it
incumbent operators throughout the world are now is quite acceptable to have to wait just a little bit for
of options to transport
ODU 1
replacing their circuit-switched based PSTN the page to download or for the page to download
n packet-switched infrastructure with a packet-switched based IP one part at a time. As we have shown in our analogy
traffic in core network technology for providing voice services. This above, both IP and MPLS offer different classes of
is usually referred to as next generation networks service.
networks. Deciding on
ODU 2
ODU 2
(NGN), which are driven by the operators’ ability to
This example of the PSTN can be generalised to all
which technology to consolidate different service networks into a single
networks, where packet-switched technology is
Trunk links use has a significant packet-switched network, thereby decreasing their
r increasingly dominating the architecture of
capex and opex.
impact on an operators’ networks.
ODU 1
Circuits (ODU 0) “Private line TDM traffic is not increasing; there are no new customers,
only some increasing bandwidth with existing customers.”
rcuit (ODU 0)
2011 2013 2016
4.4
22
3.3 OTN and MPLS
In essence, OTN is a circuit-switching technology, whereas MPLS is a technology that OTN switching is currently
introduces traffic engineering (TE) and QoS guarantees in packet-switched networks. not widely implemented by
operators in their core
network, but commercial
3.3.1 OTN
products (OTN multiplexers)
have been available for three
years and their deployment
is gathering momentum.
OTN, defined in 2003 by the International wrapper for performance management and make
Telecommunication Union (ITU-T) in its G.709 use of FEC as a means to extend the transmission
recommendation3, was created to assist the reach, which reduces the number of regenerations in
telecoms transport community to replace the legacy the network, thus reducing costs.
synchronous digital hierarchy (SDH) and synchronous
OTN multiplexing (OTH)
optical network (SONET) TDM networks and provide
a reliable and scalable transport technology for OTN was then enriched by the specification of a
multi-Gigabit/s services, driven primarily by Gigabit multiplexing hierarchy of Gigabits-per-second data
Ethernet.4 Similar to SDH, OTN is a circuit-switching rates, known as the optical transport hierarchy
technology, allocating fixed capacity on fixed circuits (OTH). Just like SDH, the OTH is based on TDM
(see the railway analogy). frames rather than packets.
It is important to differentiate two key functionalities Referring back to our analogy, OTN switching is
in OTN: equivalent to changing trains in a particular station.
Services are usually defined between two-end
• OTN encapsulation (known as digital wrapper)
stations. If the user needs to go to a destination that
• OTN multiplexing (next generation of TDM is not on route for that particular service, all
switching ) passengers will need to be grouped into a single
carriage and that carriage will have to be detached
OTN encapsulation
and re-attached to another train to reach the desired
An initial step in the creation of OTN was a digital destination. Likewise, OTN switches circuits of
wrapper to support transport of various traffic types discrete capacity called optical data unit (ODU).
in a standard ‘payload’. OTN also defined forward Figure 3.8 summarises the standardised ODU data
error correction (FEC) capabilities and nested rates.
operation, administration and management (OAM)
It should also be noted that a special ODU called
overheads to support carriers’ carrier operation.
ODU flex can be used to map packet-based services,
Referring to our railway analogy, the OTN digital
in integer number of ODU 0. For example, a 5Gbit/s
wrapper functionality is equivalent to ensuring
Ethernet service can be mapped into four distinct
that a particular train service running between
ODU 0 circuits to transport that service.
two-end stations has the necessary passenger-
carrying capacity and arrives on time at the OTN switching is currently not widely implemented
destination station, and that no passengers get by operators in their core network, but commercial
lost along the way. products (OTN multiplexers) have been available for
three years and their deployment is gathering
Currently, most wavelength division multiplexing
momentum.
(WDM) platforms use standardised OTN digital
ODU 0 1.244
ODU 1 2.499
ODU 2 10.037
ODU 3 40.319
ODU 4 104.794
3
ITU-T recommendation G.709/Y.1331 (February 2001), Interfaces for the optical transport network (OTN). Available at http://www.catr.cn/radar/itut/201007
P020100707580946344382.pdf
4
SDH was not scaled for the transportation of multi-Gigabit services. 23
WHITE PAPER
Optical transport network (OTN) and/or multi-protocol label switching (MPLS)? That is the question
3.3.2 MPLS
“This is an issue with In traditional IP networks, routing protocols disseminate information which is
interactive and time used to populate a routing table in the routers. Hence, when an IP packet is
sensitive applications: received, the router uses the packet’s destination address to perform a routing
the packet has a look-up to determine the optimum next-hop in the path towards the destination.
perishable time on it,
and it will be discarded Naturally, there may be a number of alternative In order to overcome these issues, MPLS assigns
paths towards the destination and the optimal path is short labels to network packets that describe how to
if it does not arrive at typically selected based on the shortest number of forward them through the network. This effectively
its destination within hops. This operation is repeated at each intermediate bypasses the routing protocols and establishes
router until the packet arrives to its correct express paths between any node in the network.
that time.”
destination.5 In large nodes, the routing table look-up MPLS performs ‘label’ switching instead of routing
function takes a finite amount of time, which could table look-up, which is significantly faster and
introduce some delay and jitter in packet-switched reduces the load placed on routers.
networks.
The fundamental principle of MPLS is as follows:
Also, in standard IP networks, depending on the
• the first MPLS capable router (R1) performs a
status of the network at any given time, packets from
table look-up, but instead of finding the next hop, it
the same traffic stream could experience different
finds the final destination router
delays depending on network conditions. This is an
issue with interactive and time-sensitive • a pre-determined path is then established between
applications: the packet has a perishable time on it, the source and destination routers (i.e. R1 and R9),
and it will be discarded if it does not arrive at its which is uniquely identified by an MPLS label,6
destination within that time. Furthermore, since in each segment of the network
packets are routed on a hop-by-hop basis, there is no
‘big picture’ view of network performance and an • intermediate routers (R2, R6, R7 and R5) then use
ability to respond to it quickly. Therefore, for the label to route the traffic, without needing to
example, it is possible for a particular link to become perform any additional IP look-ups
heavily congested and, yet, provided it is still in the • at the final destination MPLS router (R9), the label
path to the destination, it will continue to receive is removed and the packet is delivered via normal
traffic, exacerbating the congestion. Thus, packets IP routing.
will be queued at the corresponding router,
introducing unpredictable traffic delay and jitter and This is illustrated in Figure 3.9.
possibly even losses. Again, for real-time traffic, this MPLS allows traffic engineering to be implemented
situation would mean that the service would be in a packet-switched network, thus giving operators
impacted negatively. the ability to guarantee a minimum QoS for the
MPLS, which was standardised by the International traffic transported along each LSP.
Engineering Task Force (IETF), was introduced to
overcome all of the above issues, that is:
• provide a mechanism to alleviate the delay and
jitter associated with traditional IP routing
• provide a mechanism to ensure that all packets of
a particular traffic stream follow the same path
and, therefore, arrive in order at their destination
• provide a mechanism to guarantee a minimum
QoS in terms of bandwidth for each of the defined
paths.
24
R8 R9
Stranded c
R3
10
R2 R4
Pop
20Mbps
RESV
40 50
R6 R7 message
60 27 32
80
R1 49 R5
22
Bandwidth variation
ODU 1
guarantee a minimum
QoS for the traffic
transported along
each LSP.
ODU 2
Not enough capacity to Traffic
carry the traffic flow on flow:
that link 20Mbit/s
R8 R3 R4
10
16
ODU 1
30 30 30 R9
R2 R5 16
1:1 Mapping of
circuits
12 25 8
40 50
25
Lorry
60
80
8 Keys
R1 41 41 Empty C
12 R8
R6 R7 Used Cir
Traffic
Car Lane 2 flow:
20Mbit/s
Car Lane 1
25 Bandwidth (Mbit/s) available on each link
mergency lane
Figure 3.9: Label switched routes in an MPLS network True (
[Source: Analysys Mason, 2011] statist
All gain
flows
at
peak
value
NEW FIG.
5
Each router makes its own independent routing decisions until the final destination is reached. In our analogy, this is equivalent to a driver making the
decision on which way to go at a road junction.
6
Please note that the label may be different in a different section of the network. 25
WHITE PAPER
Optical transport network (OTN) and/or multi-protocol label switching (MPLS)? That is the question
“Today, Ethernet Ethernet has traditionally been a technology used in local area networks (LAN),
services are one of the covering distances of a few hundreds of meters (e.g. within an office or building),
largest areas of revenue but was not considered to be a carrier-grade technology, suitable for operators to
growth in packet reliably provide services over long distances
switched networks
for operators.” However, this is no longer true. The Metro Ethernet • Ethernet private line (EPL, E-Line with dedicated
Forum (MEF), a standardisation body, was created to capacity)
define carrier-grade Ethernet services and allow
• Ethernet virtual private line (EVPL, E-Line with
Ethernet to be extended beyond the LAN
shared capacity)
environment. Today, Ethernet services are one of the
largest areas of revenue growth in packet-switched • Ethernet private LAN (E-LAN with dedicated
networks for operators, as explained in Section 4.3. capacity)
The MEF has standardised the definition of two • Ethernet virtual private LAN (E-LAN with shared
families of metro Ethernet services: capacity).
• Ethernet line (E-Line) to provide point-to-point Typically, circuit-switched services can provide
connectivity. E-Line services are used to create point-to-point services with dedicated capacity.
Ethernet private-line services, Ethernet-based Therefore, circuit-switched networks can only
Internet access services, and point-to-point provide Ethernet private line services. Also, TDM
Ethernet virtual private networks (VPN). circuits are typically the same size in each
direction, whereas an operator could create an
• Ethernet LAN (E-LAN) to provide multipoint-to-
asymmetric EPL or EVPL service if a customer
multipoint (any-to-any) connectivity. E-LAN
has asymmetric traffic.
services are designed for multipoint Ethernet
VPNs and native Ethernet transparent LAN In marked contrast, packet-switched networks,
services. which offer shared capacity by default, are more
flexible and can provide the entire portfolio of
E-Line and E-LAN services are illustrated in Figure
carrier-grade Ethernet services as defined by the
3.10 and in Figure 3.11, respectively.
MEF, and in a more cost-effective manner than
Since E-Line and E-LAN services can operate over circuit-switched networks.
dedicated or shared bandwidth, four key types of
Ethernet service have emerged:
26
Since E-Line and E-LAN
“Typically, circuit-switched services can provide point-to-point7 services with dedicated services can operate
over dedicated or shared
capacity. Therefore, circuit-switched networks can only provide Ethernet private line services.” bandwidth, four key
types of Ethernet service
have emerged:
UNI
CE
UNI Carrier Ethernet
Network
CE UN
UNI Internet
ISP
UNI POP
UNI
CE
Point-to-Point
CE
N services
UNI
CE
rgin i
Carrier
Ethernet UNI Carrier Eth
Network
Network
rkk UN
UNI
CE UNI CE
200% UNI
Multipoint-to-Multipoint
Multipoint-to-Multipoint
180% CE
Point-to-Point
Relative capex efficiency
160%
b3 Figure b6
7
There are circuit-switching technologies that can provide point-to-multipoint connectivity (e.g. resilient packet ring), but they are not widely implemented
by equipment vendors. 27
160% 20
28
4 Challenges facing
operators for their
networks
This section covers the five major challenges
currently facing operators for their core networks
and how these challenges may impact their
technical strategy decisions in order to meet their
business objectives:
• How to bridge the gap between stagnating revenues and increasing network costs? (Section 4.1)
• How to cope with the increasing unpredictability of traffic patterns? (Section 4.2)
• How to maximise the use of network resources to optimise capex and opex? (Section 4.3)
• How to guarantee appropriate levels of QoS for packet-switched traffic? (Section 4.5)
29
WHITE PAPER
Optical transport network (OTN) and/or multi-protocol label switching (MPLS)? That is the question
Traffic/
capacity
Cost
Multipoint-to-Multipoint
Multipoint-to-Multipoint
Revenues
Data era
160%
350
140%
300
Total provisionned capaciity (Gbit/s)
120%
30
Relative core capex
250
100%
8
This issue is directly linked to the net-neutrality debate, but such debate falls outside the scope of this white paper.
31
WHITE PAPER
Optical transport network (OTN) and/or multi-protocol label switching (MPLS)? That is the question
“The net effect of Therefore, unless the new demand is known well in Traffic unpredictability in a packet-switched network
advance, it will be difficult, if not impossible, for the
mobility and cloud train operator to commission a new service in
Going back to our road analogy, the highway operator
computing is that does not need to know where the cars go in the first
response to a short-term demand. The same is true
place, as cars can get to their destinations by
aggregation networks for a circuit-switched network, as new circuits will
following the signs. However, there is a limit as to
have to be created to serve the new demand; if the
become less efficient.” demand changes often, this process can be quite
how much traffic can be aggregated on a road or
highway before the drivers start experiencing delays
labour-intensive and can use significant resources
on their journey due to congestion.
from the operational team, which will result in
higher opex. This is similar to packet-switched networks, which
are more flexible than circuit-switched networks in
However, there is a different way of looking at this
coping with unpredictable demand as the former
problem for circuit-switched networks. Operators
provide a natural technology to aggregate demand,
have the choice to over-dimension their circuit-
wherever it is coming from and wherever it is going.
switched network from day one (similar to adding
However, note that a packet-switched network can
extra trains knowing they will be empty most of the
only aggregate so much traffic on a trunk link of a
time). This way, any increase in traffic in any parts of
given capacity before the quality of experience for
the network could be met with minimum or no
end users starts degrading (e.g. if the network is too
network reconfiguration, thus optimising opex.
over-utilised, the quality of experience may degrade
Therefore, the operator will have to make one of
depending on how high-priority traffic is carried).
two choices:
Therefore, although the architecture of a packet-
• over-dimension its circuit-switched network from
switched network is more flexible to cope with
day one and minimise its opex at the expense of a
increased uncertainty, operators will have to weigh
very large up-front capex investment, or
the impact of degrading their QoS for particular
• make capex investments in line with the traffic traffic types (and possible churn due to customer
demand at the risk of incurring very large opex due dissatisfaction). That is where the traffic-engineering
to frequent network reconfiguration or not being capabilities of MPLS come into play. By segregating
able to carry the surge in traffic and lose business different traffic types into different classes and LSPs,
from affected customers. the operator can optimise the resources reserved for
its high-priority revenue-generating traffic and
increase the contention ratio for non-revenue
generating traffic, as mentioned in Section 4.1.
32
Traffic/
capacity
Cost
UNI
CE
Voice era Eroding margin Carrier
i
UNI
Revenues
Maximising the use of existing assets is of paramount importance for the One of the main challenges
Data
profitability ofera
any business, and the telecoms industry is no exception. for operators today is the
increasing unpredictability of
traffic in terms of the
F
In order to understand how the use of network grooming. This principle consists of consolidating amount being carried on
their networks, where it is
resources can be maximised, it is important to first traffic that is destined toFigure
a particular3.11: E-LAN
destination, so services
coming from, and where it
understand the bandwidth utilisation characteristics of that it efficiently occupies the resources. Traffic needs to go.
IP and Ethernet traffic. Figure 4.2 illustrates the grooming works quite differently in packet- and
bandwidth utilisation profile of a Gigabit Ethernet circuit-switched networks, as explained below.
Figure 4.1
file server.
Grooming of traffic in a circuit-switched network
It is clear from Figure 4.2 that the link utilisation spikes
In our railway analogy, traffic grooming in a circuit-
in short, infrequent intervals, and most of the time is
switched network is comparable to detaching
very low or even nil. Therefore, dedicating 1Gbit/s
carriages that carry people and re-attach them to a
worth of capacity for this traffic flow would be a
another train. For the train operator, this involves the
significant under-utilisation of network resources.
placement of every individual on the train to allocated
Since operators using OTN technology need to allocate
seats in the same few carriages. The same principle
fixed circuits with a fixed capacity to transport
applies to circuit-switched networks: circuits from an
160%
packet-switched traffic, they would need to allocate an
350 upstream traffic node (tributary link) have to be
ODU 0 (1.25Gbit/s), resulting in stranded capacity on 140%
mapped into the circuits of the trunk link. This is
300 that circuit that no other service can use. This
Total provisionned capaciity (Gbit/s)
250
The consolidation of resources
100% or grooming in
In order to overcome this problem, traffic can be Scenario 1: MPLS
200 Scenario 1: MPLS circuit-switched networks80% requires a 1:1 mapping
groomed by the packet-switched network at the edge
between the circuits from the tributary link and the Scenario 2: OTN
of the network and then transported onOTN
Scenario 2: an OTN circuit. 60%
150 available circuits of the trunk links. This mapping has
The problem associated with that solution is that the Scenario 3: MPLS with OTN
Scenario 3: MPLS with OTN to be performed through 40% the management system. bypass
100 OTN circuits are of fixed capacity and, should the
bypass However, if a new circuit appears on one of the
characteristics of the groomed traffic change in terms 20%
50 tributary links or if a circuit of a tributary link becomes
of capacity required, then manual intervention is
empty, that 1:1 mapping has 0% to be performed again,
needed to proportionally change the size of the circuit 1.5 1.6 1.7 1.8 1.9 2 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3
0 which, in some situations, requires manual
1.5 1.6 1.7 1.8 1.9 2 provided.
2.1 2.2 2.3 2.4This
2.5 2.6issue is further
2.7 2.8 2.9 3 discussed in Section 5.3. P/A ratio
P/A ratio intervention from the operational engineer located in
As explained in Section 3.1.2, despite the need for some the network operation centre (NOC). This constitutes a
over-provisioning, the ability of packet-switched significant burden for the operator and has an
networks to aggregate traffic and use a pool of shared associated opex. This is a rapidly escalating problem
capacity means that trunk links on packet-switched as traffic patterns are increasingly changing, as
networks typically require much less capacity than discussed in Section 4.2 of this white paper.
would be needed from an equivalent circuit-switched
Figure b2 Therefore, packet-switched technology
network. Figure
Grooming of traffic in packet-switched networks b3
optimises the use of resources available in the network. In marked contrast, grooming traffic in packet-
switched networks is more flexible because there is no
One of the key principles in trying to optimise the use
circuit defined and packets from different streams are
of resources in a core network is known as traffic
90
80
% utilisation (Gigabit Ethernet)
70
50
40
30
20
10
0
42
120
1441
1453
-
-
-
-
-
84
0
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Time (seconds)
WHITE PAPER
MPLS Switch MPLS interface WDM transponder WDM Fibre
Optical transport network (OTN) and/or multi-protocol labelOTN
switching
Switch
(MPLS)?WDM
OTN interface
That is theWDM
transponder
question
Fibre
Trunk Road
London Paris Barcelona Madrid London Paris Barcelona Madrid
Switching Switching
Layer Layer
Keys
4.3 Maximising the use of network resources to optimise capex and opex continued
WDM WDM
Layer Layer Car Lorry
WDM Bypass WDM Bypass
MPLS Switch MPLS interface WDM transponder WDM Fibre OTN Switch OTN interface WDM transponder WDM Fibre
“Nowadays, packet just aggregated onto an output consolidated stream in the trunk link to accommodate all traffic flows at their
the core network. The only intervention required by the average bandwidth simultaneously. This assumes an
networks routinely R8 operator is to increase the capacity of R9 the trunk link ideal scenario where all the peaks and troughs of the
carry voice traffic, when the incoming R3
traffic reaches a certain level. Stranded capacity
flows balance out perfectly due to statistical
as illustrated by the Therefore,
R2
reconfiguration of packet-switched
R4
10 multiplexing. This approach is likely to result
OTNinCircuit
Pop
networks will occur significantly less frequently than in congestions at some point in time due to the stochastic
adoption of NGNs by 20Mbps
circuit-switched
40 networks, resulting
50
in a lower opex
RESV nature of each traffic flow and is therefore too optimistic.
R6 message
operators.”
R7
requirement for the operational team. There is a range
A more realistic method of provisioning the trunk link
60 27 32
80
R1 of ways
49 to provision capacity in a network, R5 as
capacity is illustrated in (c) in Figure 4.4 and consists of
illustrated below in 22
Figure 4.4.
a compromiseBandwidth between method (a) and (b) by
variation
Round about
Keys Given a set of demands on a network, one can provisioning capacity for the average bandwidth for n
RSVP PATH: R1→R2→R6→R7→R4→R9
Tributary road
Train service
over-provision capacity to accommodate all traffic flows
RSVP RESV: Returns labels and reserves bandwidth
flows and provisioning peak capacity for k flows
Passenger demand at their
25 peak bandwidth
Bandwidth available on each simultaneously,
link as illustrated (assuming the total number of traffic flow is n+k). The
Rail infrastructure in (a)25 in Figure 4.4.
Label value This
returned method
via RESV
17.5Gbit/s messageof capacity provisioning
18.5Gbit/s number of trafficOTN
Tributary links
16.5Gbit/s flows provisioned
Multiplexer for peak capacity
Trunk links
Trunk Road
is equivalent to circuit-switching capacity
16.5Gbit/s
provisioning (k) will vary depending on the network characteristics,
ODU 1
and is designed to achieve absolute certainty (i.e. there the total number of flows, and the kind of services to
is enough capacity for all traffic flow to16.5Gbit/s peak at the be delivered. k can be viewed as the level of over-
Keys same time). However, this
to approach isTraffic too conservative provisioning to provide the required QoS in the
ODU 2
Not enough capacity
London
carry the traffic flow on
Paris flow: Barcelona 17Gbit/s Madrid
Lorry
Car
and is likely to result that linkin significant over-provisioning of network. Using this capacity-provisioning methodology
ODU 2
20Mbit/s
30 30 30 R9
R5 16 Paris-Madrid average traffic
as explained in London-Barcelona
Section 3.1.2). average traffic
affecting the QoS1:1ofMapping any other
circuits
of flows. It should be noted
Keys 12 25
London-Madrid average 8traffic thataverage
Barcelona- Madrid our casetraffic study in Section 5 assumes the above
Another approach
40
to capacity provisioning is illustrated
50
capacity-provisioning methodology.
25
Car Lorry
in60 (b) in Figure 4.4, 80 implying the provision of capacity on
8 Keys
Stranded capacity R1 41 41 Empty Circuits (ODU 0)
12 R8
R6 R7 NEW FIG. 5.2 Used Circuit (ODU 0)
OTN Circuit
Traffic
Car Lane 2 flow:
20Mbit/s
True (ideal)
ODU 2
statistical Remaining
All gain flows
flows at average
OTN Multiplexer Trunk links at value
peak
value
ODU 1
1:1 Mapping of
circuits 4 flows
at peak
value
Keys
ODU 2
Empty Circuits (ODU 0) “Private line TDM traffic is not increasing; there are no new customers,
only some increasing bandwidth with existing customers.”
Used Circuit (ODU 0)
2011 2013 2016
34 4 flows
11
Total Telecom (August 2011), Carrier Ethernet key to telecoms growth. Available at http://www.totaltele.com/view.aspx?ID=467030.
at peak NEW FIG. 3.7
value 12
Ibid.
The most dramatic growth in carrier Ethernet money from people who want iPhones, why can’t
services is expected to come from mobile backhaul. we make a lot of money from people who want
International market research firm Infonetics Ethernet?”.
Research expects Ethernet microwave revenues to
A recent survey11 suggested that nearly two thirds of
grow at a compound annual growth rate (CAGR) of
worldwide carriers expect to spend more on carrier
41% over the period 2011–2015.
Ethernet than on legacy wide area network (WAN)
The problem facing mobile carriers – on top of equipment during the next 18 months (with only 14%
downward price pressures – has been the surge in expecting to pay more on SONET/SDH than carrier
data traffic since the iPhone was launched in 2007, Ethernet).12
plus the fact that the smaller footprint of 3G cell
In the past five years, equipment vendors such as
sites requires more cell sites with scalable backhaul
Ciena, ADVA and ECI that have been traditionally
capabilities. Another key driver for carrier-grade
supplying transport equipment have been
Ethernet services has been video applications. For
incorporating more and more Ethernet features in
example, Netflix is now dominating bandwidth
their products to respond to the vast demand for
demand in North America and smart phones are
providing metro Ethernet services. Matt Hayes, a
pushing up the use of mobile video.
consulting systems engineer at Ciena, stated back in
Operators are rapidly responding to this increase in 2009 that “now many RFPs have carrier Ethernet in
demand by deploying packet-based infrastructure in their titles”. This shows that operators are not
their networks to support the delivery of carrier- looking for ‘transport only’ solutions anymore.
Grade Ethernet metro services:
Finally, it should be noted that Ethernet services are
• A UK operator reported in 2010 that carrier well adapted to customers’ requirements in terms of
Ethernet was its fastest-growing data service, with providing only what they need (and no more). Metro
its Ethernet business growing more than 20% Ethernet services are defined in such a way that
per annum. operators can provide additional increments of
• Tata Communications, which launched its Ethernet bandwidth to match the traffic to be increased in
global offering in 2006, has seen the service increments of bandwidth, so that customers only pay
revenues double each year till 2009. for what they need. In this way, customers can adopt
• A US operator went global by extending its Ethernet a ‘pay-as-you-go’ approach, enabling them to better
virtual private LAN service to 14 European and control their costs, which is very important in this
Asia–Pacific countries in 2009. difficult financial climate.
• BT Wholesale has been offering carrier Ethernet
There is therefore a significant opportunity for
services through its 100-node MPLS network for a
operators to generate new revenue streams, if their
few years, and is continuing to innovate in
network infrastructures are able to support the full
providing high-availability wholesale Ethernet
portfolio of carrier-grade Ethernet services, and only
products, driven by high demand from other
a flexible packet-switched network (e.g. IP/MPLS
operators. Interestingly, Tim Hubbard, head of data
network) will enable the efficient delivery of such
solutions at BT Wholesale, stated recently that10
services in an economical way.
“customers want Ethernet; Apple makes a lot of
It is very difficult to identify the optimum techno-economic solution by considering capex in isolation. For
example, lower investment may reduce the potential for new revenues, because operators with different
network architectures and/or different technologies may not be able to offer the same service portfolio.
For this reason, this study also looks at the opex implications of choosing different technical strategies,
but we intentionally do not provide any quantitative results because opex requirements are heavily
dependent on the particular business model chosen by each individual operator. Instead, we provide some
insight regarding what network opex components are affected by the technical strategy decision (see
Section 5.3).
We analyse the following three technology scenarios and identify the optimum one by comparing the
business case they each offer, using a typical long-distance reference network model:
37
OTU-v
WHITE PAPER
tch Optical
OTN interface WDMtransport
transponder network
WDM Fibre(OTN) and/or multi-protocol label switching (MPLS)? That is the question
Station A Station B Stat
witch MPLS interface WDM multiplexer
Train 1
Train 2
Core Network
As shown above, our reference topology assumes a operator. We assume that each of these nodes is
long-haul network serving London, Paris, Barcelona connected by a DWDM network, allowing a different
and Madrid. We believe that these distances are number of signals to be transmitted on the same
typical of a trans-European operator or of a US fibre, all on different wavelengths.
38
5.1.2 Traffic matrix and traffic characteristics
assumptions
Traffic matrix • Average throughput – the average traffic
bandwidth over time. We assume that the average We call the ratio between
In order to derive the capital cost of the network for
throughput is always guaranteed for each traffic peak traffic throughput and
different architecture choices we are considering a
stream in an MPLS network. average traffic throughput
packet-based14 traffic matrix, as illustrated in the peak-to-average ratio
Figure 5.2. • Peak throughput – the maximum traffic bandwidth (P/A ratio).
that can be reached by each traffic stream. We
In order to be representative of a real network, we
assume that the peak throughput is always
consider the following different types of traffic:
guaranteed for each traffic stream in an OTN
• carrier Ethernet traffic (revenue generating) architecture.
• OTT traffic, such as YouTube and peer-to-peer We call the ratio between peak traffic throughput and
(non-revenue generating). average traffic throughput the peak-to-average ratio
(P/A ratio).
In our model, the traffic matrix consists of 138
different streams of traffic.15 The source–destination A key consideration in our model is that, in a
pair for each stream of traffic is one of the following:16 circuit-switched architecture (scenario 2, described
in Section 5.1.3 below), the OTN pipes have to be
• London to Paris dimensioned for the peak throughput for each traffic
• London to Barcelona flow. Each of these traffic flow is then mapped onto
• London to Madrid dedicated ODU 0 circuit(s) equivalent to 1.25Gbit/s of
• Paris to Barcelona capacity), which is the smallest defined bandwidth
• Paris to Madrid for establishing a circuit between two end points in
• Barcelona to Madrid. an OTN. We assume that the OTN switch is ODU-flex
The exact traffic matrix is illustrated in Figure 5.2 capable.17
(and described in detail in Annex A). In marked contrast, in a packet-switched network
Please note that Figure 5.2 only represents the (scenarios 1 and 3, described in Section 5.1.3 below),
average traffic throughput between the different the network is dimensioned for the average traffic
source-destination node pairs. throughput, with enough ‘over-provisioning’ to
accommodate up to seven traffic streams to peak
Also, we assume that 20% of the traffic is OTT traffic simultaneously. In our model, the ‘over-provisioning’
and 80% is carrier Ethernet services. represents between 4% and 16% additional
Traffic characteristics bandwidth, over and above the average traffic
throughput, depending on the P/A ratio considered in
For each traffic stream, we define two parameters: the simulation.
bout
16.5Gbit/s
17Gbit/s
14
We do not model any TDM traffic as this is outside the scope of this study.
15
23 traffic streams for each source-destination pair.
Tributary links Trunk links
16
We assume bi-directional traffic on allOTN Multiplexer
links.
17
See ODU-flex definition in Section 3.3.1. 39
ODU 1
ODU 2
ODU 2
WHITE PAPER
Optical transport network (OTN) and/or multi-protocol label switching (MPLS)? That is the question
“For each scenario, Based on the above network topology and traffic matrix, we consider three
we assume that the different scenarios to assess costs and associated revenues:
underlying Layer • scenario 1: MPLS switching network traffic flow can be monitored individually). As explained
1 transmission is • scenario 2: OTN switching network (see Section 3.2), transporting Gigabit Ethernet using
provided through a • scenario 3: MPLS with OTN bypass. ‘circuits’ creates stranded bandwidth which cannot be
used (see the empty timeslots in Figure 3.5).
DWDM network, For each scenario, we assume that the underlying Layer
1 transmission is provided through a DWDM network, Also, as mentioned elsewhere in this white paper, by
which is already in which is already in place. We assume that no definition, an OTN architecture cannot provide the full
place.” regeneration is required between each of the major portfolio of carrier-grade Ethernet services, as private
nodes as this will depend on the type of DWDM virtual line21 or private virtual LAN22 can only be provided
equipment used. It should be noted that we do not by an MPLS network. Given the exponential growth of
consider the costs associated with optical line amplifiers, these services (as documented in Section 4.3), operators
because we assume that the WDM system is already in that deploy an OTN architecture with no MPLS capability
place and would be a common cost for all scenarios. risk missing out on a significant source of revenue.
Scenario 1: MPLS switching network Scenario 3: MPLS with OTN bypass
Scenario 1 involves implementing MPLS switches at Scenario 3 is essentially the same as scenario 1,
all major sites of the network, as shown in Figure 5.4. except that traffic in transit at a particular core node is
bypassed at the OTN layer, without entering the MPLS
Scenario 1 allows maximum flexibility for the operator, as it:
switch at that node (as illustrated in Figure 5.6). The
• allows the operator to quickly adapt to any
main advantage of this scenario is that traffic that is in
changes in traffic patterns without any
transit uses OTN interfaces, which are generally
reconfiguration of the network
priced at a lower level than MPLS interfaces.
• maximises the utilisation of resources in the
network, since it is in essence a packet-switched In this scenario, we differentiate between local ports
network and optimises use of bandwidth due to (required to deliver the local traffic) and line ports (that will
packet statistical multiplexing be connected to the WDM transponders on the ‘line side’).
• enables the implementation of the full carrier
However, the main issue associated with this scenario is the
Ethernet portfolio18 as defined by the MEF and
increased amount of equipment required in the network
therefore maximises the opportunity for revenues
(compared to scenarios 1 and 2). This means higher opex
• enables the implementation of QoS, with different
and capex for the network operator, which has to support
priorities for different traffic streams.19
and maintain both OTN and MPLS equipment.
Scenario 2: OTN switching network
Summary of assumptions
Scenario 2 involves implementing OTN switches at all
As shown in Figure 5.3, scenarios 1 and 3 are dimensioned
major sites of the network, with edge Ethernet
according to average traffic, with an overprovisioning of
switches20 at each site. The network architecture
4–16% of capacity to accommodate cases where up to
considered for scenario 2 is illustrated in Figure 5.5.
seven traffic streams peak at the same time. In contrast,
The main advantage of scenario 2 is that it allows scenario 2 is dimensioned according to peak traffic.
operators to:
It is interesting to note the inefficiency of the ODU 0
• provide circuit-switched (TDM) and packet-
mapping for the OTN (i.e. scenario 2). For a P/A ratio
switched (Ethernet) services with performance
of 1.5, the peak traffic to be carried is 102.5Gbit/
monitoring and maximum QoS
s×1.5=153.75Gbit/s. However, since each individual
• provide sub-wavelength aggregation for Gigabit
traffic flow will have to be mapped onto a different
Ethernet traffic.
circuit, the total capacity required in the OTN network
However, providing Ethernet services encapsulated into (scenario 2) will be 192×ODU 024, or 240Gbit/s, as
circuits means that the operator’s core network has to indicated in Figure 5.3 (see annex A for traffic matrix).
be dimensioned according to peak traffic, if every traffic This means, that on average, the utilisation of ODU 0
stream is to be provided transparently to the client circuits in the OTN will only be 63% for the traffic
purchasing these services (i.e. each Gigabit Ethernet matrix we consider.
WDM WDM
Layer Layer
MPLS Switch MPLS interface WDM transponder WDM Fibre OTN Switch OTN in
1 2 3 4 Service 1
1 2 3 4
1 2 3 4 Service 2
1 2 3 4
London Paris Barcelona Madrid London
1
Figure 5.4: Scenario 1: MPLS 2
with WDM transport 34 Service 3
1 2 34
[Source:Switching
Analysys Mason, 2011] 1 2 3 4 Service 4
1
Switching
2 3
Layer 4
Layer
1 2 3 4
1 2 3 4
WDM WDM
Layer London Paris Barcelona Madrid 1
Layer 2 3 4
WDM Bypass WDM Bypass
1 2 3 4 Circuit 1 1 2 3 4
Switching
Layer 1 2 3 4 Circuit 2
MPLS Switch MPLS interface WDM transponder WDM Fibre OTN Switch OTN in
1 2 34 Circuit 3
WDM
Layer
1 2 3 4 Circuit 4 1 1 1 1 21 21 2 32 13 232 3 3 2 1 34 2 4 433 4 4 4 4 4 Tributary road
R8 R9
London Paris Barcelona Madrid
Figure 5.5: Scenario 2: OTN with WDM transport [Source: Analysys Mason]
[Source:Switching
Analysys Mason, 2011] N/A = Not applicable
Layer R3
10
R2 R4 Keys
London Paris Barcelona Madrid Pop
WDM MPLS
Layer Switching 20Mbps Ca
Layer WDM Bypass
RESV
40 50
R6 R7 message
e OTN
OTN Switch
Switching OTN interface WDM transponder60 WDM
27Fibre 32
Layer 80
R1 49 R5
WDM 22
Layer
Keys MPLS Switch MPLS interface WDM multiplexer RSVP PATH: R1→R2→R6→R7→R4→R9
Train service R9 RSVP RESV: Returns labels and reserves bandwidth
Figure 5.6: Scenario 3: MPLS with OTN bypass and WDM transport
Passenger demand 25 Stranded
Bandwidth available on each link capacity
[Source: Analysys Mason, 2011]
R3
Rail infrastructure 25 Label value returned via RESV message
10
R4
Ethernet private line, Ethernet private LAN, Etherent virtual private line and Ethernet virtual private LAN.
18
19
Pop
It should be noted that some of the services in our traffic matrix have a guaranteed peak throughput.
20
It should be noted that edge switches are not represented in Figure 5.4 and their costs are not modelled.
21
A point-to-point service that uses shared network resources.
22
A point-to-multipoint service that uses shared network resources.
16.5 Gbit/s 41 Gbit/s
16.5
23
For P/A ratios of 1.5 and 3, respectively. RESV 16.5 Gbit/s
0 24 50 16.5 Gbit/s
18xGE traffic flows, with 500 Mbit/s peak traffic, each requiring 4xODU-0, and 120GE traffic flows requiring 1xODU-0 each.
R6 R7 message
17.5 Gbit/s 17.5 Gbit/s 18.5 Gbit/s 18.5 Gbit/s 16.5 Gbit/s 16.5 Gbit/s
80 32
Not enough capacity to Traffic
R5 carry the traffic flow on flow:
WHITE PAPER
Optical transport network (OTN) and/or multi-protocol label switching (MPLS)? That is the question
“In this case study, We assume that all equipment deployed in a scenario is costed by port. This implies
we assume that the that all equipment is reasonably full with service cards and that the key driver is the
MPLS–to-OTN port cost of the interface card of that equipment; this costing method is reasonably well
cost ratio is 1.3”. ” accepted in the telecoms industry.
In effect, the cost of the common cards and chassis equipment interface costs are normalised to this
for each type of equipment is absorbed in the cost of cost. The relative costs shown in Figure 5.7 are based
the ports. on an average discounted cost across a number of
vendors we interviewed. In this case study we
Also, in order to compare the different scenarios, we
assume that the MPLS-to-OTN port cost ratio
assume a normalised cost per port. Results from our
is 1.3.25
primary research into the cost of equipment are
summarised in Figure 5.7. Also, according to the same interviews, a long-haul
DWDM transport system 10Gbit/s port typically costs
As shown in Figure 5.7, the capital cost of an OTN
80% as much as an OTN port.
port provides the reference cost, and all other
OTN switch 1
42
5.1.5 Service pricing assumptions
For this white paper, we also surveyed the prices of carrier Ethernet services from A key result from the survey
different operators. We differentiate between two types of service: was that a shared EVPL
Gigabit Ethernet service
provided on a shared MPLS
• Ethernet private line (EPL) – point-to-point network with a guaranteed throughput of 300Mbit/s network with a guaranteed
throughput of 300Mbit/s
Ethernet circuits offered over OTN with full peak was only 25% less expensive than a dedicated EPL was only 25% less
throughput guaranteed, and Gigabit Ethernet service, providing the peak capacity expensive than a dedicated
of 1Gbit/s. It should be noted that the EVPL service EPL Gigabit Ethernet
• Ethernet virtual private line (EVPL) – point-to-point service, providing the peak
can burst up to 1Gbit/s but only 300Mbit/s of capacity
Ethernet circuits offered over MPLS with average capacity of 1Gbit/s.
is guaranteed at any moment in time.
throughput guaranteed.
The argument of many operators is that the cost of
The results are shown in Figure 5.8 below.
Ethernet service is mainly driven by the interface, and
We obtained these prices by considering inter-town providing additional capacity in the network does not
circuits in Europe. usually involve any reconfigurations. Also, in Figure
5.8, the operator price ratio between a dedicated
A key result from the survey was that a shared EVPL Gigabit Ethernet and a 10 Gigabit Ethernet is 3.
Gigabit Ethernet service provided on a shared MPLS
service
table type Capacity guaranteed (mbit/s) Peak capacity (mbit/s) relative service price
25
Section B.2 provides a cost sensitivity analysis, by varying the MPLS port cost between 1 and 2. 43
12%
80% Personnel (core network)
0% 0%
MPLS OTN MPLS + OTN MPLS OTN MPLS + OTN
“A summary of our In this section we summarise results for all three scenarios for the
Level 1 and 2 maintenance opex WDMfollowing
transponders
consumption 5.10
Level 3 and 4 maintenance opex
results is illustrated in metrics, based on the assumptions described in Section 5.1: MPLS/OTN port consumption
70% 70%
5.13
60%
• capex efficiency. 60% 5.15
WDM transponders WDM trans
20%
and 2 maintenance opex consumption 50% 18% consumptio
50%
and 4 maintenance opex MPLS/OTN port consumption MPLS/OTN
40% 40%
MPLS
MPLS
MPLS
OTN
OTN
OTN
OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
MPLS
OTN
MPLS + OTN
Provisioned Total capex Revenue Revenue per
capacity unit capex
MPLS
MPLS
OTN
OTN
OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
MPLS
MPLS
MPLS
OTN
OTN
OTN
OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
44
5.2.1 Provisioned capacity
In our case study, in order to carry the same traffic flows, an operator with an This combination of fewer
MPLS network only needs to provision 37% of the capacity compared to an OTN ports and fewer WDM
interfaces means that the
operator for a P/A ratio of 2, and 35% for a P/A ratio of 3. This is because: capex for an MPLS network
will be 47% or 42% of the
capex for an OTN network
• In an MPLS network (scenario 1), capacity is • The lowest-capacity circuit that can be used in an (for P/A ratios of 2 and 3,
provisioned according to average traffic bandwidth, OTN architecture to transport a Gigabit Ethernet respectively).
taking advantage of statistical multiplexing; and service is ODU 0 (1.25Gbit/s), which means that
the same is true for an MPLS network with some traffic will be significantly over provisioned
OTN bypass (scenario 3). However, an OTN operator (e.g. a traffic flow with a peak of 500Mbit/s will use
(scenario 2) provisions capacity according to peak a dedicated 1.25Gbit/s circuit).
traffic bandwidth, on fixed-capacity circuits.
5.2.2 Capex
Although we assume the unit cost of an MPLS port is 33% higher than an OTN
port, the lower capacity required in an MPLS network means that there are
significantly fewer ports in an MPLS network than in an OTN (for the same traffic
matrix). The lower number of MPLS ports more than compensates for the higher
unit cost of MPLS ports.
In an MPLS network with OTN bypass (scenario 3), This combination of fewer ports and fewer WDM
the ports on the line side are cheaper OTN ports interfaces means that the capex for an MPLS
(rather than MPLS ports). However, additional OTN network will be 47% or 42% of the capex for an OTN
and MPLS ports are required for the local traffic (see network (for P/A ratios of 2 and 3, respectively).
Figure 5.6), which cancels out the benefits of using
Also, the overall capex for an MPLS network with
cheaper OTN ports for bypassing the transit traffic.
OTN bypass (scenario 3) is always greater than the
Also, since less capacity needs to be provisioned in capex associated with scenario 1 (MPLS network)
an MPLS network, fewer wavelengths are required to and always lower than the capex associated with
transport the traffic between sites, which helps to scenario 2 (OTN) for all P/A ratios (see Figure B.3 in
reduce costs further (i.e. fewer WDM transponders Annex B.2).
are required in an MPLS network than in an OTN).
It is notable that WDM transponders account for
nearly 45% of the total capex in an OTN network.
45
WHITE PAPER
Optical transport network (OTN) and/or multi-protocol label switching (MPLS)? That is the question
5.2.3 Revenue
“the MPLS operators We conducted market research on what operators typically charge for Ethernet
we interviewed would services provided on an MPLS versus those provided on an OTN.
typically charge 25%
less for a Gigabit On an MPLS network, we found that Ethernet peak traffic of all traffic flows at any time. To handle
services are typically sold in such a way that only a the same traffic matrix, the MPLS operators we
Ethernet service than proportion of the peak throughput is guaranteed for interviewed would typically charge 25%26 less for a
an OTN operator, but each service, to take advantage of statistical Gigabit Ethernet service than an OTN operator, but
would only provision multiplexing. In contrast, an OTN operator will would only provision one third as much capacity. The
typically guarantee the full throughput (i.e. a full lower revenue partially offsets the tremendous cost
one third as much 1Gbit/s for Gigabit Ethernet services). Therefore, an advantage enjoyed by an MPLS operator, but overall
capacity. The lower MPLS-based operator typically offers a small it still has the potential to be much more profitable
discount because they do not provide absolute than an OTN operator, as illustrated in our analysis
revenue partially certainty that the network will be able to carry the of capex efficiency.
offsets the tremendous
cost advantage enjoyed
by an MPLS operator,
but overall it still has
the potential to be 5.2.4 Capex efficiency (revenue per invested capex)
much more profitable
than an OTN operator,
as illustrated in our
The business case for choosing the most efficient architecture is dictated by
analysis of capex
profitability. In order to assess profitability, we define capex efficiency as the relative
efficiency.”
revenue for each USD of capex invested in the network. In other words, we define
capex efficiency as revenue divided by network capex.
As shown in Figure 5.9 and Figure 5.10 earlier, the However, for an MPLS network with OTN bypass
capex efficiency for an MPLS network is significantly (scenario 3), the capex efficiency is comparable to
higher than for an OTN: for a P/A ratio of 2, the capex that of an OTN (scenario 2) for a P/A of 2. For a P/A
efficiency of an MPLS network is 58% higher than an ratio of 3, the capex efficiency for scenario 3 is 17%
OTN; and for a P/A ratio of 3 it is 81% higher. In higher than for an OTN, but still significantly lower
general, the higher the P/A, the higher the return on than for an MPLS network (scenario 2). This means
capex for an MPLS network compared to an OTN. that, overall, adding OTN ports to an MPLS network
to bypass transit traffic with cheaper ports results in
This means that for every USD of capex invested in
a significant degradation in revenue per unit capex
the network, an MPLS network will provide 58% and
compared to an MPLS network (scenario 1).
81% additional revenues for the operator compared
to an OTN (for P/A ratios of 2 and 3, respectively).
As shown in our detailed results in Annex A, the
higher the P/A ratio, the more advantageous MPLS
is over OTN.
46
5.2.5 What does the result really mean in
terms of capex and revenue?
MPLS is more suited to traffic unpredictability In marked contrast, operators that use an MPLS
Overall, our case study
network have the flexibility to dissociate bandwidth
From our interviews with operators, it is clear that illustrates that, in an OTN,
increase from provisioned capacity (and therefore capex is strongly coupled
determining the actual P/A ratio is quite challenging,
cost), leading to a much more sustainable business with an increase in network
and it is probably impossible to make an accurate capacity, which means the
model. Dissociating bandwidth from capex can, for
forecast of how the P/A ratio is going to evolve in the operator is unable to
example, be achieved by either:
short term. Traffic with random characteristics is, by dissociate traffic bandwidth
its nature, unpredictable. • implementing different classes of service for from costs.
revenue-generating and non-revenue-generating
Considering the static analysis illustrated in Figure
traffic (e.g. best effort); or
5.9 and Figure 5.10, MPLS provides a better return
on capex investment for each of the P/A ratios • increasing the contention ratio in the network, in
considered. In a real network, however, it is not order to offer an increased range of services over
possible to forecast the P/A ratio, primarily due to the same provisioned capacity (but with a reduced
the increasingly unpredictable nature of the traffic. service level agreement).
The majority of operators interviewed for this paper
Bypassing transit traffic using OTN technology will
expect the P/A ratio to increase over time, even for
not profit the MPLS operator’s business case
the highly aggregated trunk links in the core
network, but no operator is able to predict how much Perhaps a less expected result from this case study
it might increase. In this dynamic environment, is that the capex efficiency of an MPLS network with
MPLS networks provide an even better prospect than OTN bypass (scenario 3) is lower than for a native
OTN, because the higher the P/A ratio, the more MPLS network (scenario 1), but higher than that for
profitable an MPLS-based operator will be compared an OTN (scenario 2).
to an OTN-based operator (as illustrated in Figure
B.6 in Annex A). In this context, it is clear that MPLS MPLS with OTN bypass incurs higher capex than a
technology will be more suited to the unpredictable native MPLS network, principally because a
nature of future traffic. significant number of additional MPLS and OTN
ports are required for the local traffic (to
MPLS provides a more sustainable business model interconnect the OTN multiplexer and the MPLS
than OTN technology switch, as shown in Figure 5.5). The additional cost
associated with these local ports outweighs the cost
Overall, our case study illustrates that, in an OTN,
saving made by using cheaper OTN ports (compared
capex is strongly coupled with an increase in network
to MPLS ports) on the line side. Therefore, according
capacity, which means the operator is unable to
to our results, an operator that uses OTN bypass
dissociate traffic bandwidth from costs. This is a
architecture (scenario 3) will always earn less
major issue, as this business model is unsustainable
revenue per invested capex than an operator with a
if revenue stagnates, which means that operators
native MPLS network (scenario 1).
using an OTN network, will see their margins erode
as the traffic bandwidth requirement increases
(see Figure 4.1).
Please refer to the pricing benchmark presented in Section 5.1.5, where full Gigabit Ethernet (with guaranteed peak bandwidth) is charged at 25% more
25
“In this paper we Opex is always a contentious subject for operators as it very much depends on their
are evaluating opex own business model (e.g. issues such as sourcing strategy, cost of labour, capital
associated with core investment in management systems, and organisational structure).
networks, which, as
Figure 5.11 indicates, Indeed, the opex quoted in different white papers In this paper, instead of trying to identify the exact
widely vary and the cause for this variation is not opex associated with each scenario considered, we
typically represents often clear. provide some insight regarding what network opex
a significant 22% of components are affected by the decision regarding
overall opex for a technical strategy.
service provider.”
Firstly, it is important to consider how much the core network contributes to the
overall opex of an operator. A typical breakdown of network opex is provided in
Figure 5.11.
In this paper we are evaluating opex associated with In order to compare the impact of the different core
core networks, which, as Figure 5.11 indicates, architectures considered in this paper, we consider
typically represents a significant 22% of overall opex the following elements of core network opex:
for a service provider. So, any opportunity to reduce
• network maintenance (personnel)
core network opex is likely to make a substantial
difference to the overall business plan and bottom- • capacity planning and management (personnel)
line margin.
• power consumption (network and IT).
In the rest of this section, we concentrate on
network and IT and personnel opex associated with
the core network.
OTN
MPLS + OTN
Revenue per
unit capex 48
5.3.2 Maintenance opex
Depending on the business model, network maintenance is performed by different It should be noted that an
entities at various levels; for this paper, we have assumed the following operational operator with an MPLS
network using OTN bypass
model27: (scenario 3) has to maintain
two different technologies,
which will require
• Level 1 and 2 (L1 and L2) maintenance is An MPLS network that uses OTN bypass (scenario 3) additional training or the
recruitment of staff who
performed by the operator’s dedicated operational needs more ports than a native MPLS network specialise in each
teams (scenario 1), and so requires significantly more Level technology.
1(L1) and Level 2(L2) maintenance opex than a native
• Level 3 and 4 (L3 and L4) maintenance is
MPLS network (as shown in Figure 5.12 and Figure
performed by the equipment vendor, in the form of
5.13). However, the L1 and L2 opex associated with
a maintenance contract.
an MPLS network using OTN bypass (scenario 3) is
Level 1 and 2 maintenance significantly less than that needed for an OTN
(scenario 2).
As presented in Figure B.3 in Section B.2, for the
same traffic matrix, an OTN would have more than It should be noted that an operator with an MPLS
twice as many ports as an MPLS network (for a P/A network using OTN bypass (scenario 3) has to
ratio of 2, an MPLS network would have 40 MPLS maintain two different technologies, which will
ports and an OTN would have 98; for a P/A ratio of 3, require additional training or the recruitment of staff
an MPLS network would still have 40 MPLS ports but who specialise in each technology.
an OTN would need 112). Assuming that both MPLS
Level 3 and 4 maintenance
and OTN ports have the same reliability28, there will
be between 145% and 180% more faults on an OTN The annual fee charged by equipment vendors for
than on an MPLS network29. However, it would not be level 3 and 4 maintenance contracts typically
reasonable to assume that the L1 and L2 represents between 8% and 15% of the cumulative
maintenance opex is directly proportional to the equipment capex. Therefore the higher the network
number of faults in the network, as a baseline capex, the higher the maintenance fee from the
maintenance team would be required even if there equipment vendor. For the same traffic matrix, we
were very few faults in the network. However, for showed in Section that the capex for an MPLS
simplicity we assume that L1 maintenance opex is a network would be 47% or 42% of the capex for an
linear function of the number of faults in the OTN (for a P/A ratio of 2 and 3, respectively).
network, as we are just trying to show the relative Therefore, the Level 3 and 4 maintenance contracts
opex difference between an MPLS and an OTN. Using can be expected to follow the same trend.
this assumption, an operator with an MPLS network
The expected levels of Level 3 and Level 4
(scenario 1) may only incur between 36% and 41% as
maintenance opex for all three scenarios are
much opex as an operator with an OTN (scenario 2),
summarised in Figure 5.12 and Figure 5.13.
purely because of the difference in the total number
of ports.
27
The assumed operational model is just one example of those currently used in the industry.
Technically, port reliability is called mean time between failure (or MTBF).
28
MPLS
MPLS
MPLS
MPLS
OTN
OTN
OTN
OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
20% 35% 35% 26%
0%
MPLS
MPLS
MPLS
MPLS
OTN
OTN
OTN
OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
5.3.2 Maintenance opex continued Provisioned
capacity
Total capex Revenue Revenue per
unit capex
5.9
“The expected levels
of Level 3 and Level 4
maintenance opex for 100%
100%
all three scenarios are 90% 90%
12%
60% 80% 60%
MPLS + OTN
20% 35% 35% 26% Level 1 and 2 maintenance opex Total MPLS or OTN port cost Level
10% 1 and 2 maintenance o
0% Level 3 and 4 maintenance opex 0% Level 3 and 4 maintenance o
Capacity
MPLS
MPLS
MPLS
MPLS
OTN
OTN
OTN
OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
enue per
t capex
Figure 5.12: Relative Maintenance opex for a P/A
Provisioned Total capex Revenue Revenue per
ratio of 2 [Source:
capacity
Analysys Mason, 2011]
unit capex
5.12 5.13
100% 100%
90% 90%
Maintenance opex relative to OTN scenario
80% 80%
72%
50%
Relative power consumption
0% 0%
+ OTN MPLS OTN MPLS + OTN MPLS OTN MPLS + OTN
5.13
5.15
180%
160%
140% 200
50 120% 180
160
100%
140
80% 45% 158% Capex efficiency
120
60% 18% 105% Revenue 100
100% 100% 100%
40% 18% 75% 75% Total DWDM port cost 80
56%
20% 37% 37% 29% 54% Total MPLS or OTN port cost 60
5.3.3 Capacity planning and management opex
As discussed in Section 4.3, the grooming of traffic at the edge of the network is In a network, the increasing
vital in optimising resource utilisation on the network. unpredictability of where
the traffic is going to come
from and where it will be
However, because of the increasing traffic • Option 1 – over-dimensioning its circuit-switched destined for means that,
uncertainty, traffic flows are dynamically created and network from day one, in order to minimise its in a circuit-switched
environment, it will be
torn down to and from different sources and opex in return for very substantial up-front capex increasingly difficult to
destinations. In our railway analogy in Section 3.1, investment. optimise the resource of the
this is equivalent to the customer demand constantly network.
• Option 2 – making capex investments in line with
changing over time; in order to optimise the use of
the traffic demand, with the risk of incurring very
its resources, the train operator would therefore
large opex due to the need for frequent network
have to constantly adjust the number of carriages on
reconfiguration (or else being unable to carry a
existing services between stations, which would be
surge in traffic and losing business from the
very labour intensive and therefore drive up opex.
affected customers).
In a network, the increasing unpredictability of where
We note that Option 1 has been proposed by one
the traffic is going to come from and where it will be
vendor of long-haul equipment, whereby long-haul
destined for means that, in a circuit-switched
systems are pre-equipped with, say, 100Gbit/s of
environment, it will be increasingly difficult to
capacity and that capacity is provisioned and paid for
optimise the resource of the network. This is
as the demand arises. However, as stated above, this
because circuits will have to be provisioned ‘on the
very large up-front capital investment does not match
fly’, which will be labour intensive (requiring
the characteristics of the traffic that is carried on the
significant resources from the operational team) and
network in a packet-oriented model of Ethernet-
will result in higher opex.
based services. In this context, AT&T commented:
However, there is a different way of addressing this “network providers cannot economically serve their
issue for a circuit-switched network. Operators have customers by radically over-provisioning bandwidth
the option of over-dimensioning their circuit- throughout their networks to guarantee the same
switched network from day one (similar to adding low-latency, low-jitter, and low-loss performance at
extra carriages knowing they will be empty most of all times for all applications, whether those
the time initially). This approach would enable any applications are performance-sensitive or not.”30
increase in traffic in any part of the network to be
met with minimal (or no) need for network
reconfiguration, thereby minimising opex but
increasing capex. The operator must therefore
choose between:
See http://www.att.com/Common/about_us/public_policy/AT&TNet_Neutrality_Comments1_14_09.pdf.
30
51
WHITE PAPER
Optical transport network (OTN) and/or multi-protocol label switching (MPLS)? That is the question
“For the MPLS network Power consumption is been subjected to increasing scrutiny by operators due to
with OTN bypass, rising energy prices and, to a lesser extent, ecological concerns.
power consumption
is around 1.6 times For this paper, we undertook primary and secondary • The ever-increasing density integration of chips
research to benchmark the power consumption per has led to a dramatic reduction in the number of
higher than for the port for each technology considered (i.e. MPLS, OTN discrete components required. Higher-density
MPLS network (for and WDM). The power consumption per port for fully integration also means higher efficiency in the
P/A ratios of both equipped chassis is summarised in Figure 5.14. cooling of equipment, again minimising power
consumption.
2 and 3).” The results of our research for the OTN and WDM
ports were as expected, and in line with the technical Based on the consumption per port shown in Figure
equipment description sheets published by 5.14, we evaluated the overall power consumption for
equipment vendors. all three scenarios studied in this paper. The results
are shown in Figure 5.15 and Figure 5.16 (for P/A
In contrast, the results of our research into the
ratios of 2 and 3, respectively).
power consumption of MPLS ports were initially
surprising as we expected the power consumption of In Figure 5.15 and Figure 5.16, it can be seen that
MPLS ports to be higher than OTN or WDM ports, but total power consumption for the MPLS network is
it was comparable. We validated our results with significantly lower than for the OTN network, mainly
leading vendors of MPLS equipment and discovered due to the lower number of ports: MPLS network
that, the power consumption of core MPLS switches power consumption is 40% and 36% that of the MPLS
has fallen dramatically in the past three years for the network (for P/A ratios of 2 and 3, respectively).
following reasons:
For the MPLS network with OTN bypass, power
• MPLS switches use purpose-built appliances that consumption is around 1.6 times higher than for the
are optimised for MPLS switching and do not MPLS network (for P/A ratios of both 2 and 3).
perform full IP-routing functionalities. This
This is a significant result, as it indicates that the
minimises packet processing, which in turn
opex associated with power will be substantially
significantly reduces the power consumption
lower for an MPLS-based operator than for an
per bit.
OTN-based operator.
• MPLS switches use novel queuing techniques (e.g.
virtual output queuing) which reduce the on-board
memory required and minimise the number of
components a packet has to go through while in
the switch. This increases the number of packets
per second that the system can process using the
same or less power.
Assumed energy 2 2 2
consumption (Watt)
52
10%
90%
12% 12%
80% Personnel (core network) Personnel (core network)
5.10 5.10
In Figure 5.15 and Figure
5.16, it can be seen that
total power consumption
100% for the MPLS network is
% 100% 100%
significantly lower than for
% 90% 90% 90% the OTN network, mainly
due to the lower number of
80% 80% 80%
% ports: MPLS network power
50% 50%
% 0% 0% 0%
MPLS OTN MPLS
MPLS + OTN OTN MPLS + OTN MPLS OTN MPLS
MPLS + OTN OTN MPLS + OTN
Figure 5.15: Power consumption for a P/A ratio of 2 Figure 5.16: Power consumption for a P/A ratio of 3
[Source: Analysys Mason, 2011] [Source: Analysys Mason, 2011]
MPLS
OTN
OTN
OTN
56% 49%
MPLS + OTN
MPLS + OTN
MPLS + OTN
MPLS
MPLS
MPLS
OTN
OTN
OTN
OTN
MPLS
MPLS
MPLS
MPLS
OTN
OTN
OTN
OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
MPLS + OTN
mainly driven by the overall higher number of reduce their energy costs and meet their ‘green
capex Revenue
ports and cumulative capex
Revenue per
objectives’.
unit capex
• an MPLS network will require significantly less Based on the above observations, it is clear that an
network reconfiguration to reflect
Provisioned Totalchanging
capex trafficProvisioned
Revenue Revenue
capacity
perTotal capex operator
MPLS-based Revenue Revenue per
will incur significantly less
unit capex
capacity unit capex
x efficiency patterns than an OTN, resulting in a reduced opex opex than an OTN operator. This provides further
nue for capacity planning and management support for a business model where costs do not
Capex efficiency Capex efficiency
DWDM port cost increase in proportion with bandwidth requirements.
Revenue Revenue
MPLS or OTN port cost
Total DWDM port cost Total DWDM port cost
city
Total MPLS or OTN port cost Total MPLS or OTN port cost
Capacity Capacity
53
54
6 Conclusions
From our analysis, it is evident that MPLS
technology is not only more cost-effective but also
more flexible than OTN technology for providing
packet-switched services. In this respect, an MPLS
architecture will be the optimum investment for an
operator, for the following reasons:
• MPLS enables an operator to dissociate traffic bandwidth and capacity provisioning, which is key in
controlling costs and therefore breaking away from eroding margins.
• MPLS makes it possible to differentiate between different traffic types (e.g. revenue-generating versus
non-revenue-generating) and to adjust their capex spending accordingly.
• MPLS provides the ability for the operator to offer the full range of carrier Ethernet services, which
represent one of the fastest-growing markets.
• MPLS is more suited to accommodating changes in traffic patterns, which, with the advent of cloud
computing and the increase in user mobility, is becoming an increasing issue.
• MPLS will enable the operator to generate between 58% and 81% more revenues than OTN, for the
same capex investment.
• An MPLS-based operator will incur less opex compared to an OTN-based operator, which will have
significantly more ports to operate and maintain.
• MPLS architecture and technology enable an operator to significantly reduce its power consumption,
saving costs and helping it to meet its green agenda.
Therefore, a native MPLS architecture is the only one which offers the flexibility for operators to adapt
their business model in line with the changes currently being experienced in the telecoms industry, by
providing a way to differentiate revenue-generating and non-revenue-generating traffic, and to control
capex and opex.
55
WHITE PAPER
Optical transport network (OTN) and/or multi-protocol label switching (MPLS)? That is the question
This annex provides a description of all traffic flows between the different network nodes assumed for this
white paper.
It should be noted that for scenario 1 (MPLS) and scenario 3 Finally, note that the traffic matrix in Figure A.1 below has an overall
(MPLS with OTN bypass), network capacity was provisioned to P/A ratio of 1.5. The P/A ratio was varied by increasing the peak
guarantee the average traffic throughput for all traffic flows, while bandwidth for each flow, while keeping the average bandwidth
at the same time guaranteeing enough capacity for seven traffic constant.
flows peak throughput.31 For scenario 2 (OTN), network capacity
was provisioned to guarantee the peak traffic throughput for all
traffic flows simultaneously.
56 31
flows 1, 14, 30,60, 84, 103, 133
Annex A Traffic matrix continued -
Figure A.1: Traffic matrix for an overall P/A ratio of 1.5 [Source: Analysys Mason]
57
Annex A Traffic matrix continued -
Figure A.1: Traffic matrix for an overall P/A ratio of 1.5 [Source: Analysys Mason]
58
Annex A Traffic matrix continued -
Figure A.1: Traffic matrix for an overall P/A ratio of 1.5 [Source: Analysys Mason]
59
capacity
Cost
Revenues
WHITE PAPER
Optical transport network (OTN) and/or multi-protocol label switching (MPLS)? That is the question
Data era
This annex provides the detailed results obtained in our case study. It is laid out as follows:
• Section B.2 covers the core network capex associated with the 100
Scenario 3: MPLS with OTN
bypass
different scenarios to support the above provisioning of capacity
50
for each of the scenarios.
0
• Section B.3 presents our results in terms of expected revenues 1.5 1.6 1.7 1.8 1.9 2 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3
for the different architectures analysed in this white paper. P/A ratio
120
-
-
-
-
-
84
0
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
simultaneously. This capacity-provisioning approach is thought to be The capacity provisioned in an MPLS network (scenarios 1 and 3) is
conservative for packet-switched networks and is equivalent to significantly less than that provisioned in an OTN. For a P/A ratio ofTime (seconds)
methodology (c) in Figure 4.4, where k=7. 1.5 and 3.0, the total capacity provisioned in the MPLS network is
102 500Mbit/s and 115 000Mbit/s, respectively. In other words, when
Based on the above methodology, Figure B.1 shows the capacity Figure 4.2
the P/A ratio doubles, the capacity provisioned in an MPLS network
provisioned for the different networks for each of the technology
only increases by 13%. This capacity increase is due to the over-
scenarios to support the traffic matrix provided in Section 5.1.2. It
provisioning to allow up to seven traffic flows to peak at the same
should be noted that, we varied the overall traffic P/A ratio from 1.5
time.34
to 3 by increasing the peak traffic of each traffic flow (see Annex A
for more detail). Therefore, for the same traffic matrix, an MPLS network requires
between 57% and 65% less capacity to be provisioned in the network
Analysis of OTN capacity provisioning
compared with an OTN, primarily due to statistical multiplexing.
Figure B.1 shows that the total capacity provisioned on an OTN
Below we quantify how this difference in capacity provisioning
(scenario 2) is strongly correlated to the increase in peak traffic,
translates into capex.
increasing from 240 000Mbit/s to 330 000Mbit/s. It is interesting to
60
UNI
CE
i
Carrier
Ethernet UNI Carrier Ethernet
Network
rkk Network
UNI CE UN
UNI Internet
CE ISP
UNI POP
UNI
Multipoint-to-Multipoint
Multipoint-to-Multipoint CE
Point-to-Point
CE
120%
the MPLS architecture results in a 42% capex saving compared to an
100%
Scenario 1: MPLS of 3.0. Capex Efficiency for MPLS
scenario 1
80% 140%
Scenario 2: OTN A similar trend can be observed for scenario 3 (MPLS with bypassReference capex efficiency
for OTN scenario 2
60%
Scenario 3: MPLS with OTN OTN), which is between 17% and 47% cheaper than the native OTNCapex efficiency for MPLS
120%
with OTN bypass scenario 3
40% bypass
solution. Note that scenario 3 provides the added flexibility of
100%
20% transporting TDM traffic efficiently, which, depending on the
operator, can be a significant
80% source of revenues.
0%
1.5 1.6 1.7 1.8 1.9 2 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3 1.5 1.6 1.7 1.8 1.9 2 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3
P/A ratio The total number of ports for each of the considered
P/A ratio scenarios is
illustrated below in Figure B.3 for a P/A ratio of 2.0.
For scenario 1, it is important to note that, despite the fact the MPLS
ports are 1.3 times more expensive than OTN ports, the reduction in
Figure B.2: Core capex comparison for different traffic P/A ratio
the total number of ports in the MPLS scenario more than
[Source: Analysys Mason, 2011]
compensates for the unit cost differential. The cost difference due to
Figure b3 the total number of ports required for scenariosFigure
1 and 2 isb6further
exaggerated by the fact that for every OTN and MPLS port, a
B.2 Core network capex analysis wavelength has to be deployed in the DWDM network. Therefore, for
scenario 1, the network requires 40 MPLS ports and 40 WDM
Sensitivity analysis of peak-to-traffic ratio transponders, whereas for scenario 2 the network requires 98 OTN
For this exercise, we kept the MPLS-to-OTN port cost ratio constant to ports and 98 WDM transponders.
1.3 and we assumed a WDM-to-OTN interface cost ratio of 0.8. The In the case of scenario 3, bypassing the packet-switched traffic in
results are shown below in Figure B.3, where the reference cost (100%) transit on the OTN device means that fewer MPLS ports are required
is the capex of an OTN architecture (scenario 2) with a P/A ratio of 1.5. than in scenario 1 (24 MPLS ports in scenario 3 compared with 40
apacity FigureStranded
B.3 shows
capacitythat the total capex for scenarios 1 and 2 follows MPLS ports in scenario 3), but 64 OTN ports need to be added to
the same trend as the relative increase in capacity provisioning: for support both the bypass functionality and the hand-over of local
the OTN, costs increase in steps with P/A, whereas for an MPLS traffic to the MPLS switch, although the number of DWDM ports is
network, the cost remains relatively constant as the P/A ratio minimised to 40.
increases. Therefore, from an operator’s point of view, core network capex can
The cost of an MPLS network with OTN bypass architecture be better controlled using an MPLS architecture with a change in
(scenario 3) sits in between the cost of scenario 1 and scenario 2. traffic pattern (P/A ratio) compared to an OTN architecture.
1441
1453
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
me (seconds)
Scenario 2 (OTN) 98 0 98
Figure B.3: Number of 10Gbit/s interfaces required for each scenario with a traffic P/A ratio =2
[Source: Analysys Mason, 2011]
32
At the edge of the network, much higher P/A ratios are observed due to the lack of grooming, but in this white paper we only consider core networks, which are highly aggregated links.
33
Assuming ODU-flex is enabled on the network.
34
Without any over-provisioning, the capacity would remain constant at 205 000Mbit/s for varying P/A ratios.
35
Assuming an MPLS-to-OTN cost ratio of 1.3. 61
WHITE PAPER
Optical transport network (OTN) and/or multi-protocol label switching (MPLS)? That is the question
62
UNI Carrier Ethernet
Network
UN
UNI Internet
ISP
POP
UNI
Point-to-Point
CE
160%
Capex Efficiency for MPLS increasing to 181% for a P/A of 3.0. This is mainly explained by the
140%
scenario 1
fact that, in the MPLS network, capacity provisioning can be
Reference capex efficiency
for OTN scenario 2 dissociated from demand bandwidth by exploiting statistical
120% Capex efficiency for MPLS multiplexing. Also, as mentioned elsewhere in this white paper, the
with OTN bypass scenario 3
100%
lower revenues per traffic flow in the case of the MPLS network have
a smaller impact on the profitability of the network than the
80% tremendous cost savings, providing the MPLS network operator the
1.5 1.6 1.7 1.8 1.9 2 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3
P/A ratio
potential to be much more profitable than an OTN operator (because
more revenues can be generated for each USD of capex invested).
The fact that OTN operators can only provision for peak capacity and
that OTN architectures can only do so in minimum increments of
Figure B.5: Relative capex efficiency of the different network
1.25Gbit/s means that OTN operators will no longer be able to
scenarios [Source: Analysys Mason, 2011]
compete with MPLS operators because their costs and associated
pricing are higher. Therefore, we believe that OTN create an
Figure b6 unsustainable business model for packet-switched traffic.
B.4 Capex-efficiency analysis
Changes in the characteristics of traffic with the rise of Ethernet
Having calculated the costs and revenues for each of the scenarios, services, coupled with packet traffic replacing circuit traffic, is
we now analyse the relative capex efficiency for each scenario increasing the potential for higher revenues from over-subscribed
compared with the OTN scenario. We define capex efficiency as the services and creative pricing models, as well as providing the only
relative revenue for each USD of capex invested in the network. In sustainable model available to carriers today.
other words, we define capex efficiency as revenue divided by
network capex.
36
Please refer to the pricing benchmark presented in Section 5.1.5, where a full Gigabit Ethernet (with guaranteed peak bandwidth) is charged 25% more
than a Gigabit Ethernet flow with 300Mbit/s guaranteed.
37
Compared to the OTN. 63
WHITE PAPER
Optical transport network (OTN) and/or multi-protocol label switching (MPLS)? That is the question
GE Gigabit Ethernet
IP Internet protocol
64
65
Contact us
www.analysysmason.com
Edinburgh
Analysys Mason Limited Milan Washington DC
Apex 3 Analysys Limited Italia Analysys Mason Limited
95 Haymarket Terrace Via Durini 27 818 Connecticut Avenue NW
Edinburgh, EH12 5HD 20122 Milan Suite 300
Scotland Italy Washington DC 20006
UK USA
Tel: +39 02 76 31 88 34
Tel: +44 (0)845 600 5244 Fax: +39 02 36 50 41 09 Tel: (202) 331 3080
Fax: +44 (0)845 528 0760 milan@analysysmason.com Fax: (202) 331 3083
edinburgh@analysysmason.com washingtondc@analysysmason.com
66
27
www.analysysmason.com