Modeling and Analysis of Optical Backbone Networks - ProSeLex

bolivialodgeInternet et le développement Web

14 déc. 2013 (il y a 4 années et 18 jours)

350 vue(s)

Modeling and Analysis of Optical
Backbone Networks
RAJENDRAN PARTHIBAN
SUBMITTED IN TOTAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
DEPARTMENT OF ELECTRICAL AND ELECTRONIC ENGINEERING
FACULTY OF ENGINEERING
THE UNIVERSITY OF MELBOURNE
AUSTRALIA
APRIL,2004
The University of Melbourne
Australia
ABSTRACT
Modeling and Analysis of Optical
Backbone Networks
RAJENDRAN PARTHIBAN
In this thesis,we present a framework to compare and evaluate alternative
topologies and architectures for future optical backbone networks.
The most advanced form of currently deployed optical network,a point-to-
point WDMnetwork,has IP routers connected with Wavelength Division Multi-
plexed (WDM) links.Apotential bottleneck in this type of network is the capacity
of the IP routers as traffic loads increase.An alternative architecture that aims to
address this limitation is an Automatically Switched Optical Network (ASON).
The design of an optical network involves determining the number of network
elements and their interconnection topology for a given traffic demand and the
capacity constraint of each network element.We present a linear algorithm to
design an ASON.Using this algorithm,we identify the bottlenecks in an ASON,
and compare its cost to that of a point-to-point WDMnetwork.
Traffic grooming is the aggregation of lowlevel traffic flows into a higher level
traffic flow.We develop a scheme that can be used to performwaveband groom-
ing for several different topologies of an ASONthat uses single-layer multigranu-
lar Optical Cross-Connects (MG-OXCs).We also investigate howdifferent traffic
grooming schemes can be used to eliminate the bottlenecks in an ASON.
Also in this thesis,we develop a new modeling approach for ASONs,and
evaluate the cost and scalability of different architectures of point-to-point WDM
networks and ASONs as a function of traffic load.Through this modeling ap-
proach,we identify that an ASONis lower in cost than a point-to-point WDMnet-
work for lowtraffic loads.We also demonstrate that an ASONneeds IP routers to
be lower in cost for high traffic loads.We also analyzed howthe cost of an ASON
is affected by other factors such as the use of 40 Gb/s versus 10Gb/s lightpaths,
and reductions in network element cost over time.
This is to certify that
(i) the thesis comprises only my original work,
(ii) due acknowledgment has been made in the text to all other material used,
(iii) the thesis is less than 100,000 words in length,exclusive of table,maps,bib-
liographies,appendices and footnotes.
I authorize the Head of the Department of Electrical and Electronic Engineer-
ing to make or have made a copy of this thesis to any person judged to have an
acceptable reason for access to the information,i.e.,for research,study or instruc-
tion.
Signature
Date
Acknowledgments
The work described in this thesis was conducted with the assistance and support
of many people to whomI would like to express my thanks.
First,I thank God for providing me the opportunity to be one helping stone in
the great wall of human achievements.
Then,I wish to thank my parents,wife,children and friends.Without their
support and encouragement,I could not have reached this stage from a small
town in Sri Lanka.
My thanks then go to my supervisors,Prof.Rodney S.Tucker and Dr.Chris
Leckie,who gave their time and knowledge so generously.I cannot forget the
help of Dr.Jennifer Yates from AT&T and Dr.Richard Wagner from Corning In-
corporated for their valuable views on my work.
I have to thank the Department of Electrical and Electronic Engineering for
the facilities they provided,and specifically Debbie Staunton for her dedicated
help throughout my candidature.
Last but not least,I have to thank everyone at the ARC Special Research
Centre for Ultra-Broadband Information Networks for providing support for my
work.Especially,my co-students for their support,Dr.Lachlan Andrew and
Prof.Moshe Zukerman for their willingness to help always,Melinda Cartwright
for filling up the tea andsugar jars,andDr.Chandra Athaudage for his discussion
on the game of Cricket.
v
Contents
1 Introduction 1
1.1 Looking Forward.............................1
1.2 Focus of the Thesis............................3
1.2.1 Backbone Network Types....................3
1.2.2 ProblemDefinition........................7
1.3 Organization of the Thesis........................8
1.4 Contributions of the Thesis.......................9
1.5 Publications................................12
1.5.1 Journal Articles..........................12
1.5.2 Conference Papers........................12
2 Optical Backbone Networks 13
2.1 Introduction................................13
2.2 Telecommunication Networks......................14
2.3 Network Elements............................17
2.3.1 Backbone Routers.........................17
2.3.2 Optical Cross-Connects.....................18
2.4 Backbone Network Types........................22
2.4.1 IP-over-WDMIntegration....................22
2.4.2 Point-to-Point WDMNetwork.................26
2.4.3 Automatically Switched Optical Network...........29
2.4.4 IP-over-Optical Transport Network..............32
2.5 ASONNetwork Design.........................34
2.5.1 Topologies.............................34
2.5.2 Traffic Modeling.........................34
2.5.3 Design Approaches........................36
2.5.4 TDMLevel Grooming......................38
2.5.5 Waveband Grooming.......................40
2.5.6 Routing and Wavelength Assignment.............41
2.5.7 Protection and Restoration...................43
2.6 Conclusion.................................43
vii
3 Towards All-Optical Networks 47
3.1 Introduction................................47
3.2 General Network Model.........................49
3.3 Network Models.............................56
3.3.1 Network Architectures......................56
3.3.2 Preliminary Analysis.......................60
3.4 Topologies.................................64
3.5 Network Design..............................67
3.5.1 ProblemDefinition........................67
3.5.2 Integer Linear Programming ProblemFormulation.....69
3.5.3 ASON...............................70
3.5.4 IP Network............................76
3.5.5 IP Express Network.......................79
3.5.6 Fixed Network..........................81
3.6 Mathematical Model...........................81
3.6.1 Input Parameters.........................81
3.6.2 Assumptions...........................82
3.6.3 Traffic between LRs........................83
3.6.4 Initial Modeling..........................84
3.6.5 OXC Network...........................86
3.6.6 IP Network............................90
3.6.7 IP Express Network.......................92
3.6.8 Fixed Network..........................93
3.7 Search Problem..............................94
3.8 Constraints.................................95
3.8.1 Node Capacity Constraint....................95
3.8.2 Wavelength Constraint......................95
3.8.3 Wavelength Continuity Constraint...............95
3.9 Heuristic Algorithm...........................96
3.10 Geographic Mapping...........................98
3.11 Pricing Structure.............................98
3.12 Evaluation.................................100
3.13 Conclusion.................................110
4 IP Aggregation 115
4.1 Introduction................................115
4.2 Network Models.............................117
4.2.1 Network Architectures......................118
4.2.2 Preliminary Analysis.......................120
4.3 Network Design..............................123
4.4 Mathematical Model...........................124
4.4.1 Input Parameters.........................124
4.4.2 Assumptions...........................125
4.4.3 Traffic between LRs........................125
4.4.4 Initial Modeling..........................125
4.4.5 Optical IP Network........................126
4.4.6 Optical Internal IP Network...................129
4.4.7 Optical Core IP Network....................129
4.5 Search Problem..............................131
4.6 Constraints.................................132
4.7 Design and Mapping...........................132
4.8 Evaluation.................................134
4.9 Conclusion.................................154
5 Waveband Grooming 159
5.1 Introduction................................159
5.2 Network Models.............................162
5.2.1 Network Architectures......................164
5.2.2 Preliminary Analysis.......................165
5.3 Network Design..............................168
5.4 Mathematical Model...........................173
5.4.1 Input Parameters.........................173
5.4.2 Assumptions...........................174
5.4.3 Traffic between LRs........................174
5.4.4 Initial Modeling..........................175
5.4.5 Waveband Network.......................175
5.4.6 Waveband IP Network......................179
5.5 Search Problem..............................180
5.6 Heuristic Algorithm...........................181
5.6.1 Waveband Network.......................181
5.6.2 Waveband IP Network......................182
5.7 Geographical Mapping and Pricing...................184
5.8 Evaluation.................................185
5.9 Conclusion.................................195
6 Sensitivity Analysis 201
6.1 Introduction................................201
6.2 Input Parameters.............................203
6.3 Number of Ports per OXC........................203
6.3.1 Background............................203
6.3.2 Evaluation.............................205
6.4 10 Gb/s vs 40 Gb/s............................209
6.4.1 Background............................209
6.4.2 Evaluation.............................210
6.5 Manufacture Cost Reduction......................214
6.5.1 Background............................214
6.5.2 Evaluation.............................215
6.6 Port Costs.................................222
6.6.1 Background............................222
6.6.2 Evaluation.............................222
6.7 Internal Ratio...............................224
6.7.1 Background............................224
6.7.2 Evaluation.............................224
6.8 Conclusion.................................227
7 Conclusion 231
7.1 Introduction................................231
7.2 Summary of the work...........................231
7.2.1 Network Architectures......................232
7.2.2 Optical Network Design.....................235
7.3 Summary of Results............................237
7.4 Future Directions.............................241
7.4.1 Opaque OXC networks.....................241
7.4.2 Protection and Restoration...................241
7.4.3 Amplifiers,Regenerators and Performance Monitoring...242
7.4.4 Dynamic Traffic..........................243
7.4.5 Optical Burst Switching.....................243
A Required Number of Ports per OXC 259
A.1 Network Model..............................259
A.2 Port Estimation..............................260
A.3 Conclusion.................................263
B Pass-through traffic and Link Capacity 265
B.1 Internal Traffic...............................265
B.1.1 Full Mesh.............................266
B.1.2 Ring................................267
B.1.3 Partial Mesh............................270
B.1.4 Hierarchy.............................280
B.2 External Traffic..............................290
B.3 Conclusion.................................297
List of Figures
1.1 General Network Model.........................4
1.2 Backbone Network Types........................6
2.1 Telecommunication Network......................15
2.2 Optical Cross-connect Structure.....................19
2.3 Single-layer Multigranular Optical Cross-connect..........20
2.4 Multi-layer Multigranular Optical Cross-connect...........21
2.5 Traffic hierarchy defined in GMPLS...................23
2.6 Client-Server Model...........................24
2.7 Peer-to-Peer Model............................25
2.8 Augmented Model............................25
2.9 Point-to-Point WDMNetwork......................26
2.10 One node in (a) an ordinary point-to-point WDMnetwork and (b)
a special point-to-point WDMnetwork with express fibers.....28
2.11 Automatically Switched Optical Network...............29
3.1 General Network Model.........................50
3.2 Representation (in the top left) of links froman IP router......55
3.3 Representation (in the top left) of links froman OXC.........55
3.4 Fixed Network Model (showing the links fromone local router only) 57
3.5 IP Network Model............................57
3.6 IP Express Network Model.......................59
3.7 One node in (a) an IP network and (b) an IP express network....59
3.8 OXC Network Model...........................60
3.9 Traffic between two local routers....................62
3.10 Traffic between two backbone routers.................62
3.11 Traffic between two OXCs........................63
3.12 Ring topology...............................65
3.13 Full mesh topology............................65
3.14 Partial mesh topology..........................65
3.15 Hierarchical topology...........................66
3.16 Traffic flows for one node........................87
3.17 Aggregate traffic flows for one node..................87
3.18 Port allocation per node.........................88
xi
3.19 Australian Network for the OXC ring network for 10 Mb/s.....102
3.20 Overall network cost for the partial mesh topology for fixed,IP,
IP express and OXC networks (w = 40 Gb/s,v = 250,l = 1280,
l
r
= 15 Tb/s and r = 5 Tb/s).......................102
3.21 Average lightpath utilization for partial mesh topology for fixed,
IP,IP express and OXC networks (w = 40 Gb/s,v = 250,l = 1280,
l
r
= 15 Tb/s and r = 5 Tb/s).......................105
3.22 Average fiber occupancy for the partial mesh topology for fixed,
IP,IP express and OXC networks (w = 40 Gb/s,v = 250,l = 1280,
l
r
= 15 Tb/s and r = 5 Tb/s).......................105
3.23 Comparison of overall network costs for OXC and fixed networks
(w = 40 Gb/s,v = 250,l = 1280,l
r
= 15 Tb/s and r = 5 Tb/s)....107
3.24 Comparison of overall network costs for OXC and IP networks
(w = 40 Gb/s,v = 250,l = 1280,l
r
= 15 Tb/s and r = 5 Tb/s)....107
3.25 Comparison of overall network costs for OXC and IP express net-
works (w = 40 Gb/s,v = 250,l = 1280,l
r
= 15 Tb/s and r = 5 Tb/s) 108
3.26 Overall network cost for the partial mesh topology for fixed,IP,
IP express and OXC networks (w = 10 Gb/s,v = 64,l = 300,
l
r
= 2 Tb/s and r = 0.5 Tb/s)......................108
3.27 Average lightpath utilization for partial mesh topology for fixed,
IP,IP express and OXC networks (w = 10 Gb/s,v = 64,l = 300,
l
r
= 2 Tb/s and r = 0.5 Tb/s)......................109
3.28 Average fiber occupancy for the partial mesh topology for fixed,
IP,IP express and OXC networks (w = 10 Gb/s,v = 64,l = 300,
l
r
= 2 Tb/s and r = 0.5 Tb/s)......................109
3.29 Comparison of overall network costs for OXC and fixed networks
(w = 10 Gb/s,v = 64,l = 300,l
r
= 2 Tb/s and r = 0.5 Tb/s).....111
3.30 Comparison of overall network costs for OXC and IP networks
(w = 10 Gb/s,v = 64,l = 300,l
r
= 2 Tb/s and r = 0.5 Tb/s).....111
3.31 Comparison of overall network costs for OXC and IP express net-
works (w = 10 Gb/s,v = 64,l = 300,l
r
= 2 Tb/s and r = 0.5 Tb/s).112
4.1 Optical IP Network Model........................119
4.2 Optical internal IP Network Model...................119
4.3 Optical Core IP Network Model.....................120
4.4 Traffic between two OXCs in optical IP network...........121
4.5 Role of IP backbone routers in optical IP network...........127
4.6 Traffic flows for one node with IP backbone routers.........128
4.7 Overall network cost for the partial mesh topology for optical IP,
IP,IP express,and OXC networks (w = 40 Gb/s,v = 250,l = 1280,
l
r
= 15 Tb/s and r = 5 Tb/s).......................136
4.8 Average lightpath utilization for the partial mesh topology for op-
tical IP,IP,IP express,and OXC networks (w = 40 Gb/s,v = 250,
l = 1280,l
r
= 15 Tb/s and r = 5 Tb/s).................138
4.9 Average fiber occupancy for the partial mesh topology for optical
IP,IP,IP express,and OXC networks (w = 40 Gb/s,v = 250,l =
1280,l
r
= 15 Tb/s and r = 5 Tb/s)...................138
4.10 Comparison of overall network costs for optical IP and IP express
networks (w = 40 Gb/s,v = 250,l = 1280,l
r
= 15 Tb/s and r =
5 Tb/s)...................................140
4.11 Comparison of overall network costs for optical IP and OXC net-
works (w = 40 Gb/s,v = 250,l = 1280,l
r
= 15 Tb/s and r = 5 Tb/s) 140
4.12 Comparison of overall network costs for optical IPandIPnetworks
(w = 40 Gb/s,v = 250,l = 1280,l
r
= 15 Tb/s and r = 5 Tb/s)....141
4.13 Overall network cost for the partial mesh topology for optical IP,
optical internal IP and IP express networks (w = 40 Gb/s,v = 250,
l = 1280,l
r
= 15 Tb/s and r = 5 Tb/s).................143
4.14 Average lightpath utilization for the partial mesh topology for op-
tical IP,optical internal IP and IP express networks (w = 40 Gb/s,
v = 250,l = 1280,l
r
= 15 Tb/s and r = 5 Tb/s)............143
4.15 Average fiber occupancy for the partial mesh topology for optical
IP,optical internal IP and IP express networks (w = 40 Gb/s,v =
250,l = 1280,l
r
= 15 Tb/s and r = 5 Tb/s)...............144
4.16 Comparison of overall network cost for optical IP and optical in-
ternal IP networks (w = 40 Gb/s,v = 250,l = 1280,l
r
= 15 Tb/s
and r = 5 Tb/s)..............................144
4.17 Overall network cost for the hierarchical topology for optical IP,
optical core IP and IP express networks (w = 40 Gb/s,v = 250,
l = 1280,l
r
= 15 Tb/s and r = 5 Tb/s).................146
4.18 Average lightpath utilization for the hierarchical topology for op-
tical IP,optical core IP and IP express networks (w = 40 Gb/s,
v = 250,l = 1280,l
r
= 15 Tb/s and r = 5 Tb/s)............146
4.19 Average fiber occupancy for the hierarchical topology for optical
IP,optical core IP and IP express networks (w = 40 Gb/s,v = 250,
l = 1280,l
r
= 15 Tb/s and r = 5 Tb/s).................148
4.20 Overall network cost for the partial mesh topology for optical IP,
IP,IP express,and OXC networks (w = 10 Gb/s,v = 64,l = 300,
l
r
= 2 Tb/s and r = 0.5 Tb/s)......................148
4.21 Average lightpath utilization for the partial mesh topology for op-
tical IP,IP,IP express,and OXC networks (w = 10 Gb/s,v = 64,
l = 300,l
r
= 2 Tb/s and r = 0.5 Tb/s)..................149
4.22 Average fiber occupancy for the partial mesh topology for optical
IP,IP,IP express,and OXC networks (w = 10 Gb/s,v = 64,l = 300,
l
r
= 2 Tb/s and r = 0.5 Tb/s)......................149
4.23 Comparison of overall network costs for optical IP and IP express
networks (w = 10 Gb/s,v = 64,l = 300,l
r
= 2 Tb/s and r = 0.5 Tb/s)150
4.24 Comparison of overall network costs for optical IP and OXC net-
works (w = 10 Gb/s,v = 64,l = 300,l
r
= 2 Tb/s and r = 0.5 Tb/s).151
4.25 Comparison of overall network costs for optical IPandIPnetworks
(w = 10 Gb/s,v = 64,l = 300,l
r
= 2 Tb/s and r = 0.5 Tb/s).....151
4.26 Overall network cost for the partial mesh topology for optical IP,
optical internal IP and IP express networks (w = 10 Gb/s,v = 64,
l = 300,l
r
= 2 Tb/s and r = 0.5 Tb/s)..................152
4.27 Average lightpath utilization for the partial mesh topology for op-
tical IP,optical internal IP and IP express networks (w = 10 Gb/s,
v = 64,l = 300,l
r
= 2 Tb/s and r = 0.5 Tb/s).............152
4.28 Average fiber occupancy for the partial mesh topology for optical
IP,optical internal IP and IP express networks (w = 10 Gb/s,v =
64,l = 300,l
r
= 2 Tb/s and r = 0.5 Tb/s)................153
4.29 Comparison of overall network cost for optical IP and optical in-
ternal IP networks (w = 10 Gb/s,v = 64,l = 300,l
r
= 2 Tb/s and
r = 0.5 Tb/s)................................153
4.30 Overall network cost for the hierarchical topology for optical IP,
optical core IP and IP express networks (w = 10 Gb/s,v = 64,
l = 300,l
r
= 2 Tb/s and r = 0.5 Tb/s)..................154
4.31 Average lightpath utilization for the hierarchical topology for opti-
cal IP,optical core IPandIPexpress networks (w = 10 Gb/s,v = 64,
l = 300,l
r
= 2 Tb/s and r = 0.5 Tb/s)..................155
4.32 Average fiber occupancy for the hierarchical topology for optical
IP,optical core IP and IP express networks (w = 10 Gb/s,v = 64,
l = 300,l
r
= 2 Tb/s and r = 0.5 Tb/s)..................155
5.1 Representation (in the top left) of links froman OXC.........163
5.2 Links froman MG-OXC.........................163
5.3 Traffic between two nodes in a waveband network..........167
5.4 Traffic between two nodes in a waveband IP network........167
5.5 Traffic fromnode A
1
to node A
4
without grooming..........178
5.6 Traffic fromnode A
1
to node A
4
with waveband grooming.....178
5.7 Overall network cost for the hierarchical topology for the wave-
band,waveband IP,optical IP,IP,IP express,and OXC networks
(w = 40 Gb/s,v = 250,l = 1280,l
r
= 15 Tb/s and r = 5 Tb/s)....187
5.8 Average lightpath utilization for hierarchical topology for wave-
band,waveband IP,optical IP,IP,IP express,and OXC networks
(w = 40 Gb/s,v = 250,l = 1280,l
r
= 15 Tb/s and r = 5 Tb/s)....189
5.9 Average fiber occupancy for hierarchical topology for waveband,
waveband IP,optical IP,IP,IP express,and OXC networks (w =
40 Gb/s,v = 250,l = 1280,l
r
= 15 Tb/s and r = 5 Tb/s).......189
5.10 Comparison of overall network costs for waveband IP and IP ex-
press networks (w = 40 Gb/s,v = 250,l = 1280,l
r
= 15 Tb/s and
r = 5 Tb/s).................................191
5.11 Comparison of overall network costs for waveband IP and optical
IP networks (w = 40 Gb/s,v = 250,l = 1280,l
r
= 15 Tb/s and
r = 5 Tb/s).................................191
5.12 Comparison of overall network costs for waveband IP and wave-
band networks (w = 40 Gb/s,v = 250,l = 1280,l
r
= 15 Tb/s and
r = 5 Tb/s).................................192
5.13 Overall network cost for hierarchical topology for the waveband,
waveband IP,optical IP,IP,IP express,and OXC networks (w =
10 Gb/s,v = 64,l = 300,l
r
= 2 Tb/s and r = 0.5 Tb/s)........194
5.14 Average lightpath utilization for hierarchical topology for wave-
band,waveband IP,optical IP,IP,IP express,and OXC networks
(w = 10 Gb/s,v = 64,l = 300,l
r
= 2 Tb/s and r = 0.5 Tb/s).....194
5.15 Average fiber occupancy for hierarchical topology for waveband,
waveband IP,optical IP,IP,IP express,and OXC networks (w =
10 Gb/s,v = 64,l = 300,l
r
= 2 Tb/s and r = 0.5 Tb/s)........195
5.16 Comparison of overall network costs for waveband IP and IP ex-
press networks (w = 10 Gb/s,v = 64,l = 300,l
r
= 2 Tb/s and
r = 0.5 Tb/s)................................196
5.17 Comparison of overall network costs for waveband IP and optical
IP networks (w = 10 Gb/s,v = 64,l = 300,l
r
= 2 Tb/s and r =
0.5 Tb/s)..................................196
5.18 Comparison of overall network costs for waveband IP and wave-
band networks (w = 10 Gb/s,v = 64,l = 300,l
r
= 2 Tb/s and
r = 0.5 Tb/s)................................197
6.1 Three-stage Clos structure........................204
6.2 Overall network cost for hierarchical topology for the waveband
IP and optical IP network with different number of ports per OXC.206
6.3 Overall network cost for 1 Gb/s for the waveband IP and optical
IP network with different number of ports per OXC.........207
6.4 Average lightpath utilization for hierarchical topology for wave-
band IP and optical IP network with different number of ports per
OXC.....................................208
6.5 Average fiber occupancy for hierarchical topology for the wave-
band IP and optical IP network with different number of ports per
OXC.....................................208
6.6 Overall network cost for hierarchical topology in the waveband IP
network for different γ values......................211
6.7 Average lightpath utilization for hierarchical topology in the wave-
band IP network for different γ values.................212
6.8 Average fiber occupancy for hierarchical topology in the waveband
IP network for different γ values....................212
6.9 Overall network cost for the waveband IP network for 10 Gb/s
(γ = 3.36) and 40 Gb/s networks....................213
6.10 Overall network cost for partial mesh topology for the fixed,IP,IP
express and OXC networks for β = 2..................215
6.11 Overall network cost for partial mesh topology for the optical IP,
IP,IP express and OXC networks for β = 2..............216
6.12 Overall network cost for hierarchical topology for the waveband,
waveband IP,optical IP,IP,IP express and OXC networks for β = 2 217
6.13 Overall network cost for partial mesh topology for the fixed,IP,IP
express and OXC networks for β = 5..................217
6.14 Overall network cost for partial mesh topology for the optical IP,
IP,IP express and OXC networks for β = 5..............218
6.15 Overall network cost for hierarchical topology for the waveband,
waveband IP,optical IP,IP,IP express and OXC networks for β = 5 218
6.16 Overall network cost for partial mesh topology for the fixed,IP,IP
express and OXC networks for β = 10.................219
6.17 Overall network cost for partial mesh topology for the optical IP,
IP,IP express and OXC networks for β = 10..............220
6.18 Overall network cost for hierarchical topology for the waveband,
waveband IP,optical IP,IP,IP express and OXC networks for β = 10 220
6.19 Overall network cost for hierarchical topology for the waveband
IP network for varying β values.....................221
6.20 Overall network cost for hierarchical topology for the waveband
and waveband IP networks for varying α...............223
6.21 Overall network cost for the wavebandandwaveband IP networks
for varying α for 1 Gb/s.........................223
6.22 Overall network cost for hierarchical topology in the waveband IP
network for varying traffic mixture...................225
6.23 Average lightpath utilization for hierarchical topology in the wave-
band IP network for varying traffic mixture..............226
6.24 Average fiber occupancy for hierarchical topology in waveband IP
network for varying traffic mixture...................226
6.25 Overall network cost for 1 Gb/s access rate in the waveband IP
network for all topologies........................227
A.1 General Network Model.........................259
A.2 Aggregate traffic flows for one node..................261
A.3 Port allocation per node.........................262
B.1 Appendix B Overview..........................266
B.2 Traffic paths generated by a node in a ring topology.........267
B.3 Add,drop,and total traffic for an edge node in a ring topology..269
B.4 Partial mesh topology..........................270
B.5 Example paths of add traffic generated by an edge node.......271
B.6 Paths of add traffic generated by an edge node............271
B.7 Example drop traffic paths of an edge node..............272
B.8 Drop traffic paths of an edge node...................273
B.9 Example pass-through traffic paths of an edge node.........274
B.10 Pass-through traffic paths of an edge node...............275
B.11 Link capacity in links around an edge node..............277
B.12 Hierarchical topology with one core node (Type 1)..........281
B.13 Hierarchical topology with more than one core node (Type 2)....282
B.14 Explanation of one cluster for Type 2..................282
B.15 Some example paths in a hierarchical topology with one core node 283
B.16 Some example paths in a hierarchical topology with more than one
core node..................................284
B.17 Some example paths in a hierarchical topology within a cluster..286
B.18 Non-gateway nodes between two gateway nodes..........291
B.19 Example external traffic paths......................292
B.20 External traffic link capacities for an odd z...............295
B.21 External traffic link capacities for an even z..............295
List of Tables
3.1 Number of network elements and the maximumcapacity of a net-
work element...............................68
3.2 ILP formulation variables for an OXC network............72
3.3 ILP formulation variables for an IP network..............77
3.4 Internal pass-through traffic for different topologies.........91
3.5 External pass-through traffic.......................92
3.6 Pricing Structure (All prices estimated in US dollars).........99
4.2 Traffic arriving at and leaving a core node...............130
5.1 ILP formulation variables for a waveband network.........170
5.2 Pricing Structure (All prices estimated in US dollars).........185
7.1 Summary of Network Types and Features...............234
B.1 Traffic pattern adopting shortest hop path in a hierarchical topology 285
B.2 Traffic pattern counted twice in a cluster in a hierarchical topology 287
xix
Acronyms
AON All-Optical Network
ASON Automatically Switched Optical Network
BR IP Backbone Router
LR Local Router
MG-OXC Multigranular Optical Cross-Connect
OEO Optical–to–Electrical–to–Optical conversion
OOO Optical–to–Optical–to–Optical propagation
OXC Optical Cross-Connect
POP Point of Presence
TDM Time Division Multiplexing
WDM Wavelength Division Multiplexing
WGON Waveband Groomed Optical Network
Page xxii
List of Variables
Variable Name Description Unit
T Total traffic in the network lightpaths or b/s
P Number of customers –
ρ Average access rate per customer b/s
w Data-rate per lightpath b/s
v Maximumnumber of lightpaths in a fiber lightpaths
P
b
Blocking probability at the LRs %
ψ
TDM
Extra capacity allocated for inefficiencies in mul-
tiplexing TDMframes

ψ
net
Extra capacity allocated for network inefficien-
cies

ψ
bwuse
Efficiency of the use of bandwidth –
ϕ Performance margin above average access rate –
a Effective ratio above average access rate –
S Number of local routers –
n Number of edge nodes –
c Number of core nodes –
n
g
Number of gateway nodes –
z Number of non-gateway nodes between two
closest gateway nodes

Continued on the next page
Page xxiv
Variable Name Description Unit
k Number of links from each edge node to oppo-
site edge nodes in a partial mesh

k Number of edge nodes connecting a core node –
l Total number of ports in an optical cross-connect ports
m Number of ports in an optical cross-connect allo-
cated for add/drop traffic in an edge node
ports
q Number of ports in an optical cross-connect allo-
cated for pass-through traffic in an edge node
ports
l
r
Maximumcapacity of a backbone router b/s
m
r
Capacity in a backbone router allocated for
add/drop traffic in an edge node
ports
q
r
Capacity in a backbone router allocated for pass-
through traffic in an edge node
ports
r Maximumcapacity of a local router b/s
p
int
Ratio of internal traffic to the total traffic %
p
ext
Ratio of external traffic to the total traffic %
t
int
Traffic between any two local routers lightpaths
t
ext
Incoming/outgoing traffic to/froma local router lightpaths
T
loc
Intra-nodal traffic within an edge node lightpaths
T
int
Traffic between two edge nodes lightpaths
T
ext
Incoming/outgoing traffic to/from an edge
node
lightpaths
γ Ratio of 40 Gb/s interface cost to 10 Gb/s inter-
face cost

β Rate at which the cost of all network elements
decrease for a tenfold increase in the access rate

α Ratio of electronic port cost to all-optical port
cost

Rajendran Parthiban Page xxv
Page xxvi
CHAPTER ONE
INTRODUCTION
Optical network technology offers the possibility of increasing the core
network rates by several order of magnitude (∼4) andat the same time
decreases the cost per bit by approximately the same amount
Vincent W.S.Chan &J.L.LoCicero [1]
1.1 Looking Forward
On the 7
th
of March 2000,at the exhibition organized as part of the Optical Fiber
Communication (OFC) conference in Baltimore,US,a Californian startup com-
pany called Xros Inc.unveiled the first all-Optical Cross-Connect (all-OXC) that
could scale up to 1152 pairs of input and output ports.This event created huge
excitement among industry and research circles.It indicated a bright future for
All-Optical Networks (AONs) [2].In the following days,Xros Inc.was bought by
Nortel for about US$3.25 billion.This transaction was equal to paying each of the
startup’s 90 employees about US$36 million [3].All this excitement ended two
years later when Nortel decided not to bring the all-OXCto market.It announced
that it has ceased work on the cross-connect [4].The reason,according to Nortel,
was due to “dramatic changes in market conditions” [4].The question that now
arises is whether the “market conditions” can be swung back in favor of AONs.
For this to happen,researchers have to establish a clear economical justification
for building an AONover the current optical transmission infrastructure.
The most advanced formof optical network that is currently in use is known
Page 2 Chapter 1.INTRODUCTION
as a point-to-point WDMnetwork,which,has IP routers connected with point-
to-point WDMfibers [5,6,7].A major drawback of this network architecture is
that all changes to the configuration of the network must be implemented manu-
ally.This means that any attempt to reconfigure network in response to customer
requests or changes in traffic patterns will take approximately a day [8].Carriers
want the switching of data to be faster and more flexible than is possible with a
point-to-point WDM network [9,10,11].An alternative approach to the prob-
lemof faster and more flexible configuration is a proposed architecture called an
Automatically Switched Optical Network (ASON) [12].An ASONcan be imple-
mented as an AON,using Optical Cross-Connects (OXCs) to switch connections
in the optical domain.Although ASONs hold the promise of fast network re-
configuration,carriers are unlikely to deploy an ASONunless it can be shown to
have a clear cost benefit over a point-to-point WDMnetwork.
If we want to determine the cost of an ASON,we need to dimension the net-
work in terms of the number of network elements of each type that are used in
the network.For a given traffic demand and the capacity constraint of each net-
work element,an important problemis how we can calculate the number of network
elements in an ASON.
Unlike a point-to-point WDMnetwork,an ASONdoes not require expensive
optical–to–electrical–to–optical (OEO) ports for switching [13].This cost is used
to claimthat the cost of an ASONis lower than the cost of a point-to-point WDM
network [14].However,if there is any other factor that is more important than
the switching cost in the cost comparison of these networks,then this claimis not
valid.This observation leads to the following questions:
• Are there any cost factors that are more important than the switching cost?
• Howcan we identify such a factor?
Many claims [1,15,16] have been made about optical networks having a sub-
linear cost increase for increasing traffic demand (for example,the quote at the
beginning of this chapter).However,these claims need to be carefully justified
before carriers will be willing to invest in all-optical networks.In particular:
Rajendran Parthiban Page 3
• Is this true for all types of optical networks,such as a point-to-point WDM
network or an ASON?
• If there are some optical network architectures that do not exhibit a sub-
linear cost increase,can we modify these architectures to achieve this sub-
linear cost increase?
In this thesis,we answer these questions and gain insights on howfuture opti-
cal networks can be built in order to achieve a cost benefit over the existing optical
networks.
1.2 Focus of the Thesis
In this section,we explain the background of this thesis,and define the problem
we address.
1.2.1 Backbone Network Types
A telecommunication network is generally divided into three sub-networks - ac-
cess,metro and backbone [17].Wagner et al.[18] estimate that the traffic flows in
the metro and backbone networks will have similar average levels in the future,
based on their observation of current traffic loads.Consequently,they argue that
from an economical point of view,it makes sense to merge metro and backbone
networks.So,it is very likely for a future telecommunication network to have
two sub-networks - backbone and access - only.We call a backbone network with
fiber links an optical network.Ageneral model of an optical network is shown in
Figure 1.1.
In Figure 1.1,data fromthe customers reach the backbone network through lo-
cal routers.The local routers exchange data using lightpaths,which are provided
by the backbone optical network.These lightpaths formthe internal traffic of the
optical network.Some nodes of the network are directly connected to the local
routers.We call these nodes edge nodes.The local routers can also communicate
with networks of other carriers,or external networks,through selected nodes in
Page 4 Chapter 1.INTRODUCTION
Figure 1.1:General network model
the optical network.The traffic associated with this communication is called ex-
ternal traffic.Some nodes of the backbone network are directly connected to local
routers.The traffic to or from a local router may traverse several nodes in the
backbone network.Pass-through traffic is the amount of traffic that goes through
a node whose source or destination is not the node.Some nodes are not directly
connected to local routers,and are used in the network to manage pass-through
traffic.We call these nodes core nodes.
In the most advanced form of the present optical network,the nodes have
IP backbone routers and the links have WDMfibers [5].This network is known
as a point-to-point WDMnetwork [5,6,7].This network has three main disad-
vantages.The first disadvantage is the bottleneck in the capacity of electronic
switching in IP routers.For an increase in traffic,the WDMlink capacity can be
vastly expanded,whereas the same is not true for the switching capacity of IP
routers [19,20,21,22].The second disadvantage is the high cost of switching
[14].The switching cost can be reduced by using express links.Lightpaths in
these links avoid using the ports of intermediate routers for switching [23].The
third disadvantage of a point-to-point WDMnetwork is that it takes a long time
to reconfigure lightpaths for changes in traffic [8].
Rajendran Parthiban Page 5
An Automatically Switched Optical Network (ASON) provides faster switch-
ing of lightpaths than a point-to-point WDMnetwork.It has OXCs in the nodes
of the optical network.It uses signaling mechanisms that enable each network
element to identify the updated network topology and automatically reconfig-
ure the lightpaths [8].It has the potential to eliminate the capacity bottleneck
of network nodes caused by electronic processing in IP backbone routers,thus
providing scalability which allows for continuous traffic growth and network ex-
pansion [24].Despite having these advantages,there is a critical drawback in an
ASON.Traffic froma local router to an ASONis generally at a much lower level
granularity than a lightpath [25].For this reason,IP backbone routers might have
to be included in an ASONto aggregate the traffic fromlocal routers,so that it is
close to the granularity level of lightpaths.An ASONwith IP backbone routers is
known as an IP-over-Optical Transport Network (IP-over-OTN) [5].
As traffic demand increases,the number of lightpaths in an ASON will also
increase.This means that each OXC will have to switch a large number of light-
paths.However the number of ports that can be switchedin an OXChas an upper
limit.In ordinary OXCs,a lightpath is switched using a single port.When a large
number of lightpaths need to be switched in an OXC,the number of ports in the
OXC may become a bottleneck.In these circumstances,two approaches can be
used to improve scalability.The first approach is to use the ports in OXCs more
efficiently.Specifically,we can aggregate lightpaths into a waveband,or wave-
bands into a fiber,and then switch a waveband or a fiber using a single port in
an OXC.For this purpose,we use multigranular OXCs (MG-OXCs) rather than
ordinary OXCs.In an MG-OXC,we can switch a group of lightpaths using a sin-
gle port.This enables an ASONto manage a large number of lightpaths or a high
traffic volume.We call an ASON capable of performing waveband grooming
a Waveband Groomed Optical Network (WGON).The second approach to deal
with limited number of ports in an OXC is to increase the effective number of
ports,by interconnecting several OXCs together to forma multi-stage OXC with
a large number of effective ports.
We summarize in Figure 1.2 the different types of backbone networks that we
Page 6 Chapter 1.INTRODUCTION
consider in this thesis.We have already described the difference between a point-
to-point WDMnetwork and an ASON.Each of these networks have several vari-
ations.Apoint-to-point WDMnetwork can be ordinary or with express links.An
ASON can also be a hybrid of the different variations we discussed above.For
example,an ASON can be an IP-over-OTN and a WGON at the same time.In
other words,it can use IP backbone routers and be capable of performing wave-
band grooming.Thus,an ASONcan be ordinary,an IP-over-OTN,a WGONor a
combination of an IP-over-OTNand a WGON.If an ordinary ASONor a WGON
uses all-OXCs,then it can be all-optical.
Figure 1.2:Backbone Network Types
As backbone traffic increases,telecommunication carriers must decide what
type of backbone network to deploy to meet this growing demand.In order to be
able to make this decision,we need to address the following challenges:
• For a given traffic scenario,how can we find the number of network el-
ements and their interconnection topology for these network types,while
taking into consideration any constraints on the use of each type of network
Rajendran Parthiban Page 7
elements?
• How can we evaluate these alternatives to identify the most cost-effective
network type for future?
The focus of this thesis revolves around these key questions.We model each
network type,and evaluate their cost benefits and scalability for a range of traffic
scenarios.By means of this evaluation,we gain insights on how a future optical
network can be built in order to achieve a sub-linear cost increase with increasing
traffic demand.
1.2.2 ProblemDefinition
In order to compare different optical backbone network types for a given traffic
scenario,we first need to develop a method for designing each network type.
The inputs to this design process are the number of customers connected to the
network and their traffic demands.Using these parameters,the number of local
routers required at the edge of the network can be calculated.These local routers
are to be interconnected by the optical backbone network.We consider the differ-
ent network types in Figure 1.2 for this network.The design of each network type
involves calculating the number of network elements and their interconnection
topology.This design problem is NP-complete [26].We simplify this problem
by considering only symmetric topologies.The symmetry requirement of these
topologies are later removed by including the geographical constraints.We de-
sign the network for a static traffic demand fromthe customers.Optical impair-
ments such as signal degradation are ignored in the design process.Our focus in
this thesis is only on designing a non-protected optical backbone network.
Using this design process,we compare the network types in Figure 1.2 for a
range of traffic demands fromcustomers.In particular,we are interested in what
is the impact of increasing traffic demands on the following range of issues.
The first issue is identifying the bottlenecks in an ASON.We mentionedearlier
that a point-to-point WDMnetwork has a bottleneck in the capacity of its routers
Page 8 Chapter 1.INTRODUCTION
for increasing traffic demand.One may ask:“Is there any such bottleneck in an
ASON?” We answer this question in this thesis.
The second issue we address is identifying a cost-effective network.We men-
tioned earlier that claims [1,15,16] have been made that optical networks have a
sub-linear increase in cost for increasing traffic demand.We examine the network
types in Figure 1.2 and identify the network type(s) that have this feature.
We mentioned earlier that an ASONcan be an IP-over-OTNand/or a WGON.
Using these network types,we address a thirdissue for increasing traffic demand,
which is determining the effect of IP aggregation and waveband grooming in the
design of an ASON.
The final issue,perhaps the most important issue,we address in this thesis,
is developing a general framework to compare the performance of different net-
work types.
1.3 Organization of the Thesis
The chapters in thesis are organized as follows:
Chapter 2:Optical Backbone Networks surveys the relevant literature.This
chapter explains the key issues affecting the design of a point-to-point WDMnet-
work and an ASON.This chapter also examines the approaches used for design,
routing and wavelength assignment,and the basic concepts of protection and
restoration.
Chapter 3:Towards All-Optical Networks explains the modeling of three net-
work types in Figure 1.2 – (1) an ordinary point-to-point WDM network,(2) a
point-to-point WDMnetwork with express links,and (3) an ordinary all-optical
ASON.This chapter then defines the network design problem,and explains the
design procedure for these network types.Then a comparison is made of the
network cost and network element utilization of these network types.Using this
comparison,this chapter identifies the key bottlenecks in an ASON.
Rajendran Parthiban Page 9
Chapter 4:IP Aggregation investigates three different approaches to IP aggre-
gation in an ASON,and howthe IP aggregation affects the overall network cost.
In other words,we investigate three different sub-types of an IP-over-OTN,the
(4)th network type shown in Figure 1.2.After explaining the design procedure
for these network sub-types,a comparison is made with the network sub-types in
Chapter 3 in terms of network cost andnetwork element utilization.This compar-
ison shows howIP aggregation helps to eliminate some bottlenecks in an ASON.
Chapter 5:WavebandGrooming examines the role of wavebandgrooming and
investigates its effect in the design of an ASON.This chapter shows how wave-
band grooming can be included in an ASON,and describes two types of wave-
band groomed ASON or WGON (these two types are numbered (5) and (6) in
Figure 1.2).After specifying the modeling and the design procedure of these
WGON types,this chapter compares the performance of all the network types
from Chapter 3 to 5.This comparison demonstrates how waveband grooming
and IP aggregation can be used to eliminate the bottlenecks in an ASON and
make an ASONmore cost-effective than a point-to-point WDMnetwork.
Chapter 6:Sensitivity Analysis explores how the conclusions drawn fromthe
previous chapters would be affected by changing various parameters,such as the
number of ports per OXC,the bit rate of a lightpath,and the cost of network
elements.
Chapter 7:Conclusion summarizes the main results of this thesis.It also out-
lines the future work that can be undertaken in this area.
1.4 Contributions of the Thesis
We consider the problem of how to design optical backbone networks that can
interconnect a set of local routers.We model the network types in Figure 1.2,
which are variations of a point-to-point WDMnetwork and an ASON.We then
Page 10 Chapter 1.INTRODUCTION
develop design procedures for each of these network types.Using these design
procedures as a basis,we present a framework to compare the performance of
these network types.Specifically,we make the following contributions in this
thesis.The relevant chapters in which the contributions are made,are given in
each case.
1.We derive mathematical models for the network types in Figure 1.2,and
evaluate their performance for a national backbone network (Chapter 3,
Chapter 4,Chapter 5 and Chapter 7)
2.We develop a computationally efficient algorithmto design an ASONback-
bone (Chapter 3)
3.We identify two key bottlenecks in an ASON(Chapter 3).They are:
(a) The quadratic increase in the number of lightpaths for increasing traffic
demand
(b) The limitation in the number of ports per OXCfor high traffic demands
4.We determine the effect of IP aggregation in the design of an ASON for
increasing traffic demand (Chapter 4).We showthat:
(a) The cost-effective point in an ASONto performIP aggregation is in the
network’s edge nodes
(b) The IP aggregation eliminates the bottleneck in the number of light-
paths
(c) The IP aggregation reduces the cost of an ASON by a factor of 5 for
high traffic demands
5.We develop a scheme for incorporating waveband grooming using single-
layer MG-OXCs in an ASON,and investigate the effect of this scheme in the
design of an ASON as a function of traffic demand (Chapter 5).We show
that the waveband grooming:
(a) Eliminates the bottleneck in the number of ports per OXC
Rajendran Parthiban Page 11
(b) Reduces the cost of an ASONby a factor of 4 for high traffic demands
6.Develop a framework for comparing the overall network cost and utiliza-
tion of network resources for these network types (Chapter 3,Chapter 4,
Chapter 5 and Chapter 7).Using this framework,
(a) We identify topologies that provide low cost networks for designing
an ASON(Chapter 5)
(b) We show that an ASON has a greater cost advantage over a point-to-
point WDMnetwork for lowtraffic demands (Chapter 3)
(c) We showthat an ASONcosts less than a point-to-point WDMnetwork
with express links for high traffic demands only if the IP aggregation
and waveband grooming are performed (Chapter 5)
(d) We demonstrate that an ASON can achieve a sub-linear cost increase
with increasing traffic demand only if the IP aggregation and wave-
band grooming are performed (Chapter 5)
(e) We investigate the cost trade-off in increasing the bit rate of lightpaths
for an ASONas a function of traffic demand (Chapter 6)
(f) We investigate the effect of increasing the number of ports in an OXC
for an ASONas a function of traffic demand (Chapter 6)
(g) We investigate the effect of the ratio of internal to external network
traffic in an ASONfor increasing traffic demand (Chapter 6)
7.We derive analytical expressions that would help in the design of an ASON.
Specifically,
(a) We estimate the required number of ports per OXC for a given traffic
demand in an ASON(Appendix A)
(b) We derive expressions for pass-through traffic for regular symmetric
topologies (Appendix B)
Page 12 Chapter 1.INTRODUCTION
1.5 Publications
During the course of this project,a number of publications have been made which
are based on the work presented in this thesis.They are listed here for reference.
1.5.1 Journal Articles
• R.Parthiban,R.S.Tucker and C.Leckie,”Waveband Grooming and IP Ag-
gregation in the Design of Optical Networks”,Journal of Lightwave Technol-
ogy,vol.21,no.11,pp.2476-2488,November 2003 [27].
1.5.2 Conference Papers
• R.Parthiban,R.S.Tucker and C.Leckie,”Waveband Grooming and IP Inte-
gration in Optical Ring Networks” in Proc.Conference on the Optical Internet
(COIN) 2003,Melbourne,Australia,July 2003,p.103-106 [28].
• R.Parthiban,R.S.Tucker and C.Leckie,”Determining the Required Num-
ber of Ports in an Optical Cross-Connects in GMPLS networks” in Proc.Con-
ference on the Optical Internet (COIN) 2003,Melbourne,Australia,July 2003,
p.523-526 [29].
CHAPTER TWO
OPTICAL BACKBONE NETWORKS
When the Internet was invented in 1969 or even a decade after that,
no one dreamed of the enormous popularity of the Internet today
L.Kleinrock [30,31]
2.1 Introduction
Over the last decade,the number of users of the Internet has increased tremen-
dously.This has increased the traffic in backbone networks all over the world.
In the US,Internet traffic growth is estimated to be more than 100% per annum
[32,33,34],regardless of conditions in the financial market [35].With the growth
in the number of Internet hosts,the same trend is expected in the rest of the world
as well [36,37,38].In addition,other newservices such as video on demand,elec-
tronic distance learning and grid computing applications will increase the traffic
in broadband networks [38,39,40].If these services gain wide acceptance over
the next several years,then these services coupled with new access technologies
will produce a dramatic increase in traffic.Consequently,carriers will be required
to increase the bandwidth of their backbone networks to account for this increase
in traffic [41].
Optical fiber technologies are the prime choices for transmitting the enormous
bandwidth required by the traffic growth in backbone networks.For example,
Wavelength Division Multiplexing (WDM) can dramatically increase the capacity
of fiber links by allowing many signals at different wavelengths to be multiplexed
Page 14 Chapter 2.OPTICAL BACKBONE NETWORKS
onto a single fiber.Fine wavelength spacing of these signals is known as Dense
Wavelength Division Multiplexing (DWDM).Asingle fiber with DWDMhas the
ability to accommodate terabits of information [42,43].
This chapter is organized as follows.In the next section,we introduce the
basic concepts of a general telecommunication network.Then we focus on the
backbone part of this network – specifically,on the backbone network that uses
optical fibers.In Section 2.3,we explain in detail the network elements that are
used in the present backbone network or are likely to be used in a future network.
In the subsequent section,we describe the different network types that are pos-
sible candidates for future optical networks.In Section 2.5,we summarize the
key issues for the design of an important candidate future broadband networks,
namely,the Automatically Switched Optical Network (ASON).We also explain
the approaches used for design,routing and wavelength assignment,and the ba-
sic concepts of protection and restoration.
2.2 Telecommunication Networks
In this section,we explain some basic concepts of a general telecommunication
network.Then,we focus on backbone networks and investigate some key issues
that affect the design of this network.Finally,we define the terminology that we
use in this chapter.
A telecommunication network is generally divided into three sub-networks -
access,metro and backbone [17].We show a schematic of a telecommunication
network in Figure 2.1.
In Figure 2.1,data from the customers reach the backbone network through
the access,and then the metro network.The access network is typically a ring or
star network.The links in an access network can be fibers,copper wires or even
wireless.Fromthe access network,the traffic is transported to the metro network
through local routers.These routers are also known as access routers [45].Since
each local router manages traffic froma relatively small community of customers
who are connected to the local access router,its capacity is relatively low [45].
Rajendran Parthiban Page 15
Figure 2.1:Telecommunication Network (based on [44])
Along with local routers,the metro network also has a metro point-of-presence
(POP).The roles of the metro POP are (1) adding/dropping traffic to/from the
metro network,(2) switching and grooming traffic from/to local routers,(3) serv-
ing as a pass-through hub for backbone network traffic,(4) taking appropriate
measures in the event of link failures,and (5) having facilities to differentiate
services [44].The access traffic is groomed into a higher capacity by the metro
POP and is transferred to the backbone network [44].In the current broadband
backbone networks,the metro POP and the nodes in the backbone network have
IP backbone routers associated with them [5].The key purpose of these routers
is to route IP packets to their correct destination in the backbone network [44].
Since the backbone routers have to manage traffic frommultiple local routers,the
required capacity of these routers are higher than the local routers.In the most
advancedtelecommunication network,links in backbone and metro networks are
generally fibers running WDMtechnology.We call a backbone network with fiber
links an optical network.
Historically,the traffic in telecommunication networks was predominantly
voice.Eight years ago,the average level of traffic in the links of a metro net-
Page 16 Chapter 2.OPTICAL BACKBONE NETWORKS
work,or the traffic granularity in the metro networks was between 625 Mb/s to
2.5 Gb/s [18].In the same year,the traffic granularity in backbone networks was
from 2.5 Gb/s to 10 Gb/s.For this reason,grooming of traffic from the metro
network to the backbone network was necessary.Today,the dominant traffic in
telecommunication network is from data due to Internet traffic.Wagner et al.
[18] estimate that in 2006,the traffic flows in the metro and backbone network
will have the same granularity.They support this estimate with observations of
traffic granularity in current backbone links.Based on this estimate,they argue
that froman economical point of view,it makes sense to merge metro and back-
bone networks.So,it is very likely for a future telecommunication network to
have only two sub-networks - backbone and access.For the rest of this chapter,
we focus our survey on backbone optical networks.
As we mentioned earlier,typical telecommunication network today is based
on IP backbone router nodes that are connected by fiber links using WDMtech-
nology.In this network,IP routers are responsible for routing and traffic man-
agement,and the fiber links provide capacity.In other words,all the intelligence
resides in the IP routers.Many authors,eg.Dixit and Ye [46],believe that in
future optical networks some or all intelligence will be transferred to the optical
layer [46].We describe the network elements that can facilitate this transfer in
Section 2.3,and discuss key issues that affect the transfer in Section 2.4.
Network carriers will only introduce new technology in their backbone net-
works if this new technology can reduce their capital and operational expendi-
tures [47].Moreover,future networks should be able to provide different services
to increase revenues,and be scalable with increasing traffic demand from cus-
tomers [48].In Section 2.4,we investigate different backbone network types and
identify factors that are likely to affect the cost of these network types.As we will
see in Section 2.4,a comprehensive model that can compare today’s and future network
types is missing in the literature.
Before we move to other sections,let us first define several key terms we use in
this chapter.We need to distinguish between a lightpath and a wavelength.Alight-
path is a unidirectional optical communication channel with a fixed bandwidth
Rajendran Parthiban Page 17
and electronic terminations at its two ends.In the absence of wavelength con-
verters,each lightpath is assigned a single wavelength throughout the network.
We also need to distinguish between the physical and logical topologies [49].
A physical topology is a set of nodes and the fiber-optic links connecting them.
According to a given traffic demand,lightpaths traverse this physical topology.
The set of all lightpaths and their end nodes constitute the logical topology or vir-
tual topology [49].This is the topology seen by the IP layer and may look totally
different fromthe physical topology [50].
2.3 Network Elements
In the previous section,we explained the structure of a general telecommunica-
tion network and turned our focus to backbone optical networks.In this section,
we describe the different network elements and technologies that are currently in
use or have the potential to be part of a future optical network.We explain in the
next section how these elements and technologies are used in different network
types.
2.3.1 Backbone Routers
We explained the purpose of backbone routers in Section 2.2.In this sub-section,
we discuss howhigh capacity backbone routers are made.
There are three main methods reported in the literature for constructing high
capacity routers [51,52,53].Each method aims to connect low capacity routers
together to build a high capacity router,where the capacity of routers can be char-
acterized in terms of their number of ports.The first method is from Cisco Sys-
tems Inc.According to this method,low capacity routers are connected using a
full mesh to make a high capacity router.In this method,when the number of
low capacity routers are increased,the overall number of usable ports increases,
reaches a maximum value,and then decreases [52].Hence,with this method,
there is an upper bound for the maximumoverall capacity that can be achieved.
Page 18 Chapter 2.OPTICAL BACKBONE NETWORKS
Tzang and Mandviwalla [53] uses a constant number of ports in each low
capacity router to interconnect with other low capacity routers.Then the low
capacity routers are interconnected using horizontal and vertical crossbars,which
use the allocated ports.In this method,each low capacity router loses a certain
capacity for interconnection.Moreover,a high speed crossbar fabric is required
to connect lowcapacity routers,which may not be practical.
The third method is from Dally [51],which is adopted by Avici Systems.
According to this method,low capacity routers are constructed using a three-
dimensional torus topology.The number of interconnection ports to create this
torus topology is not counted in the capacity of a low capacity router.If mul-
tiple low capacity routers are connected together preserving the torus topology
structure,Dally [51] shows that high capacity routers can be constructed without
loosing any capacity ports for interconnection.
In our opinion,the thirdmethodis better than the other methods because of its
ability to achieve high capacity without loosing any capacity ports for intercon-
nection.However,one cannot keep on increasing the capacity by interconnecting
routers using this method forever,because of the speed limitation in managing
look up tables in an IP backbone router [54].
2.3.2 Optical Cross-Connects
Optical cross-connects (OXCs) are network elements used for switching light-
paths [55].Specifically,an OXC provides switching functionality between l in-
put ports and l output ports.OXCs that can switch 1280 ports with each port or
lightpath operating at 40Gb/s have already been experimentally demonstrated
[56].There are a number of technologies used for switching.Some of them are
optical Micro-Electrical-Mechanical-Systems (MEMS),thermal,and electro-optic
based systems [57].In this sub-section,we describe several types of OXCs and
their internal structure.
The general structure of an OXC is shown in Figure 2.2.In this figure,thin
arrows represent lightpaths,gray shaded areas represent fibers,and trapezoids
Rajendran Parthiban Page 19
represent multiplexors or demultiplexors.In an OXC,the lightpaths in a fiber
are separated by a demultiplexor (DeMUX).Each lightpath arrives at a separate
input port in the switching matrix,and then it is switched to the respective output
port.Finally,the lightpaths fromthe output ports are multiplexed together using
a multiplexor (MUX).
Figure 2.2:Optical Cross-connect Structure
There are two types of OXCs - opaque and transparent [58,59].Opaque OXCs
convert the incoming lighpath into an electrical signal for switching;the electri-
cal signal is converted back to an (optical) lightpath at an output port.Hence,an
opaque OXC uses an optical–to–electrical–to–optical (OEO) interface to switch
lightpaths.An opaque OXC can provide a variety of functions such as optical
regeneration,reshaping and retiming (3R),and wavelength conversion.A major
disadvantage of an opaque OXCis its expensive OEOinterface [60,61].Transpar-
ent OXCs (or all-optical OXCs),with their optical–to–optical–to–optical (OOO)
ports,switch the signals in the optical domain and eliminate OEOinterfaces.Sig-
nals (or lightpaths) switched through OOOports have flexibility in bit-rate,mod-
ulation format,and protocol [62,63].By eliminating the OEO interfaces,we can
gain significant cost benefits as well [60,61].Transparent OXCs are less expensive
than opaque OXCs,consume less power,and generate less heat [48].However,
they cannot performfunctions such as 3Rregeneration or wavelength conversion.
Page 20 Chapter 2.OPTICAL BACKBONE NETWORKS
Ordinary transparent OXCs allowa lightpath to be switched as a single entity
(i.e.using a single port).The structure of this OXC can be modified to switch
a group of lightpaths using a single port.An OXC with this ability (waveband
grooming ability) is known as a multi-granular OXC (MG-OXC).Multigranular
OXCs are used to switch traffic at multiple levels of granularity,namely,at the
fiber,waveband or lightpath levels.Awaveband is formed by grouping multiple
lightpaths that have the same source or destination (we discuss about wavebands
in detail in Section 2.5.5).In MG-OXCs,each level of granularity (i.e.lightpath,
waveband or fiber) is switched using a single port [64].Two types of MG-OXC
– single-layer MG-OXC [65] and multi-layer MG-OXC [66,67,59] – are proposed
in the literature.We show these types in Figures 2.3 and 2.4.In these figures,
thin arrows represent lightpaths,thick arrows represent wavebands,gray shaded
areas represent fibers,and trapezoids represent MUXs or DeMUXs.
Figure 2.3:Single-layer Multigranular Optical Cross-connect
In a single-layer MG-OXC (see Figure 2.3),there are three logical regions for
switching fibers (FXC),wavebands (WXC) and lightpaths (LXC) [65].Some fibers
pass-through the OXC through the FXC region,while other pre-configured fibers
are separated into wavebands or lightpaths.Some of these wavebands pass-
through the WXC region,while other pre-configured wavebands are separated
into lightpaths [65].The LXC region switches lightpaths.
Rajendran Parthiban Page 21
Figure 2.4:Multi-layer Multigranular Optical Cross-connect
In a multi-layer MG-OXC (see Figure 2.4),the three regions - FXC,WXC and
LXC - are physically separated.The fibers in this MG-OXC pass-through the FXC
region [67].If any one of the fibers need to be switched in terms of wavebands or
lightpaths,then it is separated into wavebands and sent to the WXCregion.Some
wavebands pass-through the WXC region.If any one of the wavebands need to
be switched in terms of lightpaths,then it is separated into lightpaths and sent to
the LXC region [59].
Cao et al.[64] argue that when there are multiple fibers,wavebands and light-
paths,the multi-layer MG-OXC is more complicated than the single-layer one.
They point out another advantage of the single-layer MG-OXC,which is better
signal quality,because all lightpaths go through only one switching fabric.In
contrast,for the multi-layer MG-OXC,some lightpaths may go through two or
three switching regions (i.e.FXC,WXC,and LXC).They also describe a disad-
vantage in using the single-layer MG-OXC.Only pre-configured fibers or wave-
bands can be dropped in the single-layer MG-OXC,while the remaining wave-
bands can pass-through it.In contrast,the multi-layer MG-OXC does not have
this constraint and switching is more flexible [64].A recent work by Cao et al.
[68] suggests that the blocking inside a single-layer MG-OXC is more than the
multi-layer MG-OXC,if the traffic demand and the number of ports are the same.
Page 22 Chapter 2.OPTICAL BACKBONE NETWORKS
The main advantage in using MG-OXCs over the ordinary OXCs is their abil-
ity to minimize the required number of ports.This is not surprising,because
MG-OXCs switch a group of lightpaths using a single port,whereas ordinary
OXCs switch each lightpath using a separate port.In order to accommodate the
traffic in a network,the single-layer MG-OXC requires the least number of ports
and the multi-layer MG-OXC requires fewer ports than the ordinary OXC [64].
There are other network elements used in optical networks such as amplifiers
[69] and regenerators [70,71,72].Similarly,there are other switching elements
which could be part of a future optical network,such as an optical burst switch
(OBS) that can be used to switch bursts of IP packets [73].We will not go into the
details of these network elements in this literature survey.
2.4 Backbone Network Types
In the previous section,we described different network elements and technolo-
gies used in an optical network.In this section,we start with describing the need
for integration of IP and WDMlayers.Then,we focus on different optical back-
bone network types that use the network elements described in the previous sec-
tion.We first explain the currently used network types and then move towards
future network types.
2.4.1 IP-over-WDMIntegration
In this section,we outline why the focus of a future optical network is going to be
on two layers - IP and WDM.We also discuss the different aspects that need to be
considered when integrating these two layers.
At present,WDM deployment is mostly point-to-point,and uses the syn-
chronous optical network (SONET or SDH) as the standard layer for interfac-
ing to the higher layers of the protocol stack.On the other hand,IP routers and
Asynchronous Transfer Mode (ATM) switches in nodes use the IP and ATMlay-
ers.The IP layer routes the majority of the traffic transported by carriers,whereas
Rajendran Parthiban Page 23
WDMserves as a bandwidth rich link layer.If we can combine the IP and WDM
layers and remove all the other layers,this will simplify the management of the
layers tremendously [74].Generalized Mutliprotocol Label Switching (GMPLS)
was proposed in the literature as a novel approach to integrate the IP and WDM
layers [75].GMPLS is a control plane,which provides generalized labels to traffic
flows with different switching types [76,77,78].In GMPLS,the traffic hierarchy
is defined from lowest to highest level as packets,layer 2 frames,TDM frames,
lightpaths,wavebands,and fibers (See Figure 2.5).
Figure 2.5:Traffic hierarchy defined in GMPLS ([79])
For each level of this hierarchy,labels are also defined [78].Nodes in opti-
cal networks equipped with GMPLS signaling are defined as Generalized Label
Switching Routers (G-LSRs).The IP layer manages the packet,layer 2 frame,
TDM frame and lightpath hierarchy levels,whereas the optical layer manages
the lightpath,waveband,and fiber hierarchy levels.If the optical and IP layers
are brought together using GMPLS,then they can jointly manage all levels in the
traffic hierarchy [79].
Another aspect in the integration of the IP and WDMlayers is how different
information such as data path provisioning is exchanged between these two lay-
ers.Three models are proposed for this purpose - client-server,peer-to-peer and
Page 24 Chapter 2.OPTICAL BACKBONE NETWORKS
augmented [80,81,82].We explain these models briefly below.
In the client-server model,IP routers act as clients and the optical layer as
server (See Figure 2.6).The IP routers in this model make a circuit-switched
lightpath request to an Optical Connection Controller (OCC) through the opti-
cal user–to–network–interface (O-UNI).The OCC then asks the optical network
to provide a path.After a path is found,the client packets are carried through the
optical network with OXCs.Different OCCs communicate with each other us-
ing the optical network–to–network–interface (O-NNI).In this model,no routing
information is exchanged between the IP layer and the optical layer [46].
Figure 2.6:Client-Server Model ([46])
In the peer-to-peer model,IP routers and OXCs in the optical domain are
treated equally (See Figure 2.7).Both IP and optical (or WDM) layer have the
same control plane.These layers exchange complete information about routing
[46].
The augmented model falls between the client-server and peer-to-peer models
(See Figure 2.8).The client (IP) layer and the optical (WDM) layers have separate
control planes.Both of these layers exchange limited routing information [46].
Since different carriers will be reluctant to exchange information regarding
their network topology and routes,the client-server model is likely to be pre-
ferred initially,with a possible evolution towards the augmented model [46].
Rajendran Parthiban Page 25
Figure 2.7:Peer-to-Peer Model ([46])
Figure 2.8:Augmented Model ([46])
Page 26 Chapter 2.OPTICAL BACKBONE NETWORKS
However,it is unlikely for the peer-to-peer model to be used,because to imple-
ment this model,different carriers have to share complete information about their
topologies and routing paths.
2.4.2 Point-to-Point WDMNetwork
Today’s optical networks are typically point-to-point WDMnetworks.These net-
works connect IP backbone routers using point-to-point WDMfibers [5,6,7].We
can visualize this network as having IP backbone router nodes and WDM fiber
links.We showa basic schematic of this network in Figure 2.9.In this figure,ar-
rows represent lightpaths,gray lines represent fibers,black lines represent links,
and trapezoids represent MUXs or DeMUXs.In Figure 2.9,we showthe details of
fibers,lightpaths,MUXs and DeMUXs for one IP backbone router.These details
are omitted for others for clarity.
Figure 2.9:Point-to-Point WDMNetwork
In Figure 2.9,local routers deliver data to/from customers.This data is con-
verted to lightpaths and sent to a backbone router.At each backbone router,
lightpaths are converted to Time Division Multiplexed (TDM) frames consisting
Rajendran Parthiban Page 27
of packets.After the route and destination of these packets are identified,they
are then converted back to lightpaths before being sent through the WDMlinks.
TDMframes are processed in backbone routers in the electronic domain.
A point-to-point WDMnetwork has three main disadvantages.The first dis-
advantage is the bottleneck in the capacity of electronic IP routers.For an increase
in traffic,the WDMlink capacity can be vastly expanded,whereas the same is not
true for the switching capacity of IP routers [19,20,21,22].There are experimen-
tally demonstrated cases [42,43],in which the WDM link capacity reaches up
to 100 Tb/s.As for the capacity of an IP router,it is expected to be limited to
around 10 Tb/s for the foreseeable future due to the computational overhead in
managing the routing look-up tables [54,83].
The second disadvantage is the high cost of switching.Electronic switching
is arguably more expensive than optical switching [14].This can be understood
by looking at the amount of pass-through traffic in the backbone network.For
example,the North American backbone network has an average of 72% pass-
through traffic through its nodes [13].In an electronic router,all of this pass-
through traffic needs to be switched electronically,whereas in optical switching
elements,such OXCs,this traffic can be switched all-optically.So,the general
belief is that if the pass-through traffic can be sent through OXCs,then switching
cost can be reduced [13].
Clemente and Ferraris [8] point out a third disadvantage of a point-to-point
WDM network,namely the long delays in reconfiguring lightpaths for new or
changing traffic demands.Carriers typically take approximately a day to recon-
figure lightpaths in response to a specific customer request or a change in traffic
pattern [8].Nadon [13] explains the reconfiguration process as follows.When a
customer wants a new traffic request set up,the customer contacts the network
provider with the request.Then the network provider asks the network carrier
for a newroute for the request.The network carrier investigates the topology of
the network and manually looks for a route for the new request.After the route
is found,the network is reconfigured by reconfiguring all IP routers of the route
and the new connection becomes operational.If one can automate this process,
Page 28 Chapter 2.OPTICAL BACKBONE NETWORKS
this would make the switching process significantly faster.
It is possible in a point-to-point WDMnetworks to use express links [23].This
eliminates the second disadvantage of an ordinary point-to-point WDMnetwork.
An ordinary point-to-point WDMnetwork is expensive due to the switching cost
of sending all the pass-through traffic through IP routers [14].Doverspike et al.
[23] have proposed that pass-through traffic can be sent through express links.
Figure 2.10 shows the difference between an ordinary point-to-point WDMnet-
work node and one with express links (Thin arrows represent lightpaths,gray
shaded areas represent fibers,and trapezoids represent MUXs/DeMUXs).In an
ordinary point-to-point WDMnetwork,all traffic - add,drop and pass-through -
use the ports of the IP backbone router as shown in Figure 2.10(a).In a point-to-
point WDMnetwork with express links,add and drop traffic uses the ports of the
backbone router and pass-through traffic does not.Identifying the express links
requires the knowledge of the pass-through traffic in the network.Knowing the
amount of pass-through traffic in advance is not possible for a general topology
and traffic scenario.For this reason,pre-provisioning the express links is a hard
problem[23].
Figure 2.10:One node in (a) an ordinary point-to-point WDMnetwork and (b) a
special point-to-point WDMnetwork with express fibers
Rajendran Parthiban Page 29
In a point-to-point WDMnetwork,pass-through traffic is sent through back-
bone routers.Hence,we get the flexibility of switching lightpaths in the elec-
tronic domain manually,when the traffic characteristics vary.When the express
links are used for pass-through traffic as in Figure 2.10(b),we lose the flexibility in
terms of reconfiguration when traffic from customers varies.In a point-to-point
WDMnetwork,we pay more in terms of ports in backbone routers to achieve this
flexibility.
2.4.3 Automatically Switched Optical Network
Carriers want the switching of lightpaths to be faster and more flexible to changes
in traffic than a point-to-point WDM network [9,10,11].If we replace IP back-
bone routers in a point-to-point WDMnetwork with OXCs as the main switching
(network) element,then this optical network is expected to provide these features
[12].The optical network with OXCs is known as an Automatically Switched Op-
tical Network (ASON) [9,10,11,12].This network can be visualized as having
optical cross-connect nodes with WDMfiber links.We showa basic schematic of
this network in Figure 2.11 ([12]).
Figure 2.11:Automatically Switched Optical Network
Page 30 Chapter 2.OPTICAL BACKBONE NETWORKS
In Figure 2.11,the lightpaths from local routers are sent to an OXC through
a WDM link.Extra information such as routing preferences are sent through
the user–to–network–interface (UNI).Some of these OXCs can have backbone
routers connected to them.We discus the aggregation of traffic using OXCs and
backbone routers in Section 2.4.4.In an ASON,different optical subnets may be
owned by different carriers [46].Different optical subnets have links between
them.Through these links,both user data and routing information can be sent
between carriers.The interface which carries this information is known as the
external network–to–network–interface (E-NNI) [12].
In an ASON,lightpaths must satisfy what is known as the lightpath continuity
constraint [84].This constraint has two aspects.Each lightpath has a wavelength
assignedto it.The first aspect of the constraint is that a lightpath must be assigned
the same wavelength on all the links in its route.The second aspect is that any
two lightpaths sharing an optical fiber must use two different wavelengths to
avoid any wavelength conflicts.
An ASON provides faster switching of lightpaths by using signaling mecha-
nisms that enable each network element to identify the updated network topol-
ogy and automatically reconfigure the lightpaths [8].Moreover,an ASONcan (1)
eliminate the capacity bottleneck of network nodes caused by electronic process-
ing in IP backbone routers,(2) provide scalability which allows for continuous
traffic growth and network expansion,and (3) deliver excellent survivability by
providing restoration in the order of 10 - 1000 milliseconds in the event of a failure
[24].
Despite having these advantages,there are some critical issues in an ASON.
We outline one of these issues below.For the huge amount of traffic that ag-
gregates in the backbone network,it is better to use high granularities such as
10 Gb/s and 40Gb/s for lightpaths to reduce the total number of lightpaths in
the network,and hence reduce the number of ports used in OXCs in an ASON
[64].On the other hand,we cannot expect customers to generate the volume of
traffic to occupy lightpaths at this level of granularity.So,we need to use traffic
grooming techniques to groomthe customer traffic at some point in the network
Rajendran Parthiban Page 31
to aggregate the traffic close to the granularity level of lightpaths [25].This raises
a number of questions:
1.Where can we do this grooming?
2.Howcan we groomwithout adding significant extra cost?
3.If we performtraffic grooming,which topology is best suited for this?
To the best of our knowledge,no reported work is available that addresses these
issues comprehensively.We mentioned earlier that if we use OXCs,as in an
ASON,the switching cost is less than the IP router nodes of a point-to-point
WDM network.When we compare the cost of these two networks,other ques-
tions also arise:
4.An ASON cost is expected to be less than a point-to-point WDM cost,be-
cause the switching cost of an ASON is less than the switching cost of a
point-to-point WDMnetwork [14].Is their any other factor that dominates
the cost comparison of these networks more than the switching cost?
5.How will an ASON cost compare with the cost of a point-to-point WDM
network with express links,which avoids the significant cost of switching the
pass-through traffic?
Doverspike et al.[23] address the fifth question above for a fixed traffic scenario.
We discuss this work in detail in Section 2.4.4.However,to the best of our knowl-
edge,no analysis has been made of these questions as we vary the level of traffic
demand that needs to be carried by the network.There are some other open is-
sues with regard to an ASON.They are:
6.For an increasing traffic demand,what is the cost trade-off between using
10 Gb/s and 40Gb/s lightpaths in an ASON?
7.If we increase the number of ports in an OXC by interconnecting multiple
OXCs,how will this affect the cost of an ASON for increasing traffic de-
mand?
Page 32 Chapter 2.OPTICAL BACKBONE NETWORKS
An ASON does not have to be all-optical.If we use an opaque OXC instead
of a transparent OXC (all-optical OXC) in this network,switching faster than a
point-to-point WDMnetworks is still possible.An opaque OXC has the ability to
provide wavelength conversion.For this reason,an ASONwith opaque OXCs is
likely to need fewer lightpaths in each link compared to an ASONwith transpar-
ent OXCs [85].If we compare the cost of an ASONwith opaque OXCs (O-ASON)
and an ASONwith transparent OXCs (T-ASON),then there are three factors that
affect the comparison.The first factor is the number of lightpaths needed in an
O-ASON as opposed to a T-ASON.The second factor is the extra cost of OEO
ports of an O-ASONover OOOports of a T-ASON.
The third factor is the ability to reduce the number of ports used in an OXC
using waveband grooming.Opaque OXCs do not have this ability because of
their OEO ports [58],whereas the transparent OXCs do.For high traffic demand
scenarios,there may be situations where the number of ports in OXCs may be-
come a bottleneck.In these circumstances,we might need waveband grooming.
For this case,we can use only a T-ASON and not an O-ASON.This is a major
disadvantage of an O-ASON.For different traffic demands,an analysis of which
of these factors dominates the others will give insights into the cost comparison
between a T-ASONand an O-ASON.To the best of our knowledge,this analysis
is missing in the literature.
2.4.4 IP-over-Optical Transport Network
An IP-over-Optical Transport Network (IP-over-OTN) is an ASON,in which the
nodes use OXCs and IP backbone routers [5].This network is presented as a
network that can save cost more than an ordinary ASON with OXCs or a point-
to-point WDM network.Sengupta et al.[5] outline some of the advantages of
an IP-over-OTNover a point-to-point WDMnetwork:(1) avoiding “expensive”
IP backbone router ports for pass-through traffic,(2) providing a “more scalable”
solution by sharing traffic between IP backbone routers and OXCs,and (3) po-
tentially achieving restoration for IP router failures in the optical layer.Based on
Rajendran Parthiban Page 33
a network analysis using “traffic data from service providers”,Sengupta et al.
[5] showthat the IP-over-OTNis more cost-effective than a point-to-point WDM
network.However,there are several important limitations to their analysis as
pointed out below.
Sengupta et al.[5] argue that IP backbone router ports are expensive for pass-
through traffic.In a point-to-point WDMnetwork with express links,this is not
true.Doverspike et al.[23] find a point-to-point WDM network with express
links to be more cost effective than an IP-over-OTN.Sengupta et al.[5] assume
that IP router failures can be totally restored using the optical layer.Many other
authors [86,87] challenge this view.This is because some degree of restoration
capacity must be provided for failures in the IP layer itself,since these failures
cannot be restored at the optical layer.
We believe a point-to-point WDM network (with or without express links)
might be more cost-effective than an IP-over-OTN for some traffic scenarios;the