energy-efficient next-generation networks (e2ngn) · energy-efficient next-generation networks...
TRANSCRIPT
Energy-Efficient Next-Generation Networks (E2NGN)
By
Pulak ChowdhuryB.S. (Bangladesh University of Engineering & Technology) 2002
M.S. (McMaster University, Canada) 2005
Dissertation
Submitted in partial satisfaction of the requirements for the degree of
Doctor of Philosophy
in
Computer Science
in the
Office of Graduate Studies
of the
University of California
Davis
Approved:
Biswanath Mukherjee, Chair
Dipak Ghosal
Xin Liu
Committee in Charge
2011
-i-
To my beloved wife, fabulous daughter, and wonderful parents. . .
-ii-
Contents
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . x
1 Introduction 1
1.1 Energy Efficiency in Core Networks . . . . . . . . . . . . . . 3
1.2 Energy Efficiency in Access Networks . . . . . . . . . . . . . 3
1.2.1 WOBAN and Energy Efficiency . . . . . . . . . . . . . 4
1.3 Research Contributions . . . . . . . . . . . . . . . . . . . . . 5
1.3.1 Energy Efficiency in Telecom Optical Networks . . . 5
1.3.2 WOBAN Prototype and Research Challenges . . . . . 6
1.3.3 Building a Green WOBAN . . . . . . . . . . . . . . . . 6
1.3.4 Energy-Efficient Mixed-Line-Rate Network Design . . 7
1.4 Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2 Energy Efficiency in Telecom Optical Networks 9
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Network Domains . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.1 Core Network . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.2 Metro Network . . . . . . . . . . . . . . . . . . . . . . 17
2.2.3 Access Network . . . . . . . . . . . . . . . . . . . . . . 20
2.3 Standardization Efforts . . . . . . . . . . . . . . . . . . . . . 21
2.4 Core Network . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.4.1 Selectively Turning Off Network Elements . . . . . . 24
2.4.2 Energy-Efficient Network Design . . . . . . . . . . . . 26
2.4.3 Energy-Efficient IP Packet Forwarding . . . . . . . . 28
2.4.4 Green Routing . . . . . . . . . . . . . . . . . . . . . . 32
2.5 Access and Metro Network . . . . . . . . . . . . . . . . . . . 35
-iii-
2.5.1 Energy Consumption Estimation . . . . . . . . . . . . 36
2.5.2 Energy-Aware Access Networks . . . . . . . . . . . . 37
2.6 Data Centers and Applications . . . . . . . . . . . . . . . . . 44
2.6.1 Data Centers . . . . . . . . . . . . . . . . . . . . . . . 44
2.6.2 Applications . . . . . . . . . . . . . . . . . . . . . . . . 48
2.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3 WOBAN Prototype Development and Research Challenges 54
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.2 Related Development Efforts . . . . . . . . . . . . . . . . . . 56
3.3 Implementing WOBAN Prototype . . . . . . . . . . . . . . . . 57
3.3.1 Resources Needed . . . . . . . . . . . . . . . . . . . . 57
3.3.2 Architecture . . . . . . . . . . . . . . . . . . . . . . . . 58
3.3.3 Distinguishing Features . . . . . . . . . . . . . . . . . 60
3.3.4 Development Procedure . . . . . . . . . . . . . . . . . 61
3.4 Experimental Illustrations . . . . . . . . . . . . . . . . . . . 65
3.4.1 Experimental Setup . . . . . . . . . . . . . . . . . . . 65
3.4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.4.3 Critical Observations . . . . . . . . . . . . . . . . . . 71
3.5 Research Challenges . . . . . . . . . . . . . . . . . . . . . . . 72
3.5.1 Layer-2 Integrated Routing . . . . . . . . . . . . . . . 73
3.5.2 TDM MAC for Wireless . . . . . . . . . . . . . . . . . . 73
3.5.3 Improve Flexibility in WOBAN Architecture . . . . . . 74
3.5.4 Hierarchical Architecture . . . . . . . . . . . . . . . . 74
3.5.5 Energy-Efficiency in WOBAN . . . . . . . . . . . . . . 74
3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4 Building a Green WOBAN 76
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
-iv-
4.3 Green WOBAN . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.3.1 WOBAN Architecture . . . . . . . . . . . . . . . . . . . 80
4.3.2 Energy-Aware WOBAN Design . . . . . . . . . . . . . 81
4.3.3 Energy-Aware Routing . . . . . . . . . . . . . . . . . . 89
4.4 Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.4.1 Traffic Modelling . . . . . . . . . . . . . . . . . . . . . 94
4.5 Illustrative Numerical Examples . . . . . . . . . . . . . . . . 96
4.5.1 MILP vs. Heuristics . . . . . . . . . . . . . . . . . . . 101
4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
5 Energy-Efficient Mixed-Line-Rate (MLR) Network Design 104
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.2 IP-over-WDM Network Architectures . . . . . . . . . . . . . . 107
5.2.1 Transparent Architecture . . . . . . . . . . . . . . . . 107
5.2.2 Translucent Architecture . . . . . . . . . . . . . . . . 108
5.2.3 Opaque Architecture . . . . . . . . . . . . . . . . . . . 109
5.3 Energy-Efficient MLR Network Model . . . . . . . . . . . . . 110
5.3.1 Transparent IoWDM Network . . . . . . . . . . . . . . 112
5.3.2 Translucent IoWDM Network . . . . . . . . . . . . . . 114
5.3.3 Opaque IoWDM Network . . . . . . . . . . . . . . . . 116
5.4 Illustrative Numerical Examples . . . . . . . . . . . . . . . . 117
5.4.1 Reach Estimation . . . . . . . . . . . . . . . . . . . . 118
5.4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
6 Conclusion 132
6.1 Summary of the Research Contributions . . . . . . . . . . . 132
6.2 Future Research Directions . . . . . . . . . . . . . . . . . . 134
6.2.1 Core Networks . . . . . . . . . . . . . . . . . . . . . . 134
6.2.2 Metro Networks . . . . . . . . . . . . . . . . . . . . . . 136
-v-
6.2.3 Access Networks . . . . . . . . . . . . . . . . . . . . . 136
-vi-
List of Figures
2.1 Energy consumption forecast of telecom networks. . . . . . 11
2.2 Telecom network hierarchy. . . . . . . . . . . . . . . . . . . 13
2.3 Core network. . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 Metro and access networks. . . . . . . . . . . . . . . . . . . 19
2.5 Parallel networks on the same fiber infrastructure. . . . . 31
2.6 Green routing with availability of renewable energy. . . . . 33
2.7 Network connectivity “Proxying”. . . . . . . . . . . . . . . . 49
2.8 Grid Computing job scheduling mechanism. . . . . . . . . 50
3.1 WOBAN prototype architecture. . . . . . . . . . . . . . . . . 59
3.2 WOBAN prototype experimental setup. . . . . . . . . . . . . 64
3.3 Data-transfer throughput. . . . . . . . . . . . . . . . . . . . 66
3.4 VoIP performance: Packet-loss rate. . . . . . . . . . . . . . . 67
3.5 VoIP performance: Jitter. . . . . . . . . . . . . . . . . . . . . 67
3.6 VoIP performance: Mean Opinion Score (MOS). . . . . . . . 68
3.7 Video streaming performance: Packet-loss rate. . . . . . . . 69
3.8 Video streaming performance: Jitter. . . . . . . . . . . . . . 69
3.9 Video streaming performance: Video quality. . . . . . . . . 70
4.1 WOBAN architecture . . . . . . . . . . . . . . . . . . . . . . . 81
4.2 Residual capacity as link weights. . . . . . . . . . . . . . . . 90
4.3 Hypothetical WOBAN deployment in Davis. . . . . . . . . . 93
4.4 Traffic profile: Ratio of active routers. . . . . . . . . . . . . . 94
4.5 Traffic profile: Average load on active routers. . . . . . . . . 95
4.6 Power savings in energy-aware WOBAN. . . . . . . . . . . . 96
4.7 Power savings vs. extra wireless power. . . . . . . . . . . . . 96
4.8 Energy-aware WOBAN performance: Average path length. . 98
4.9 Energy-aware WOBAN performance: Average path delay. . 98
-vii-
4.10Energy-aware WOBAN performance: ONU utilization. . . . 99
4.11Performance of MILP and heuristics: Average path length. 102
5.1 Transparent IoWDM architecture. . . . . . . . . . . . . . . . 108
5.2 Translucent IoWDM architecture. . . . . . . . . . . . . . . . 109
5.3 Opaque IoWDM architecture. . . . . . . . . . . . . . . . . . . 110
5.4 Cost239 topology (link lengths in km). . . . . . . . . . . . . 120
5.5 Energy cost comparison of transparent networks . . . . . . 123
5.6 Energy cost comparison of translucent networks . . . . . . 123
5.7 Energy cost comparison of opaque networks . . . . . . . . . 123
5.8 Transponder distribution in a transparent MLR network. . 128
5.9 Transponder distribution in a translucent MLR network. . 128
5.10Regenerator distribution in a translucent MLR network. . . 129
5.11SRT distribution in an opaque MLR network. . . . . . . . . 129
5.12OOT distribution in an opaque MLR network. . . . . . . . . 130
-viii-
List of Tables
2.1 Typical power values of different components. . . . . . . . . 18
2.2 Comparison of greening efforts in core networks. . . . . . . 29
2.3 Comparison of greening efforts in PON. . . . . . . . . . . . 38
3.1 WOBAN prototype components and their specifications. . 58
4.1 Energy savings vs. Low watermark. . . . . . . . . . . . . . 100
5.1 Base traffic matrix. . . . . . . . . . . . . . . . . . . . . . . . 121
5.2 Energy consumption values of network components. . . . . 122
5.3 Energy consumption of transparent networks’ components. 124
5.4 Energy consumption of translucent networks’ components. 125
5.5 Energy consumption of opaque networks’ components. . . 126
-ix-
Acknowledgments
First and foremost, I would like to thank Dr. Biswanath Mukherjee,
my mentor and supervisor, for keeping faith on me in this enduring and
persevering journey of PhD study. No words are enough to describe his
blessings on me. Someone once told me, “PhD mentor is like your father,
he will guide you through when going gets tough.” I now believe every bit
of that statement. He has always posed new set of challenges in front of
me and guided me through the process through his constant support,
wisdom, and moral boost. His pursuit of perfection, dedication, and
attitude of taking nothing-but-the-best have refined myself as a better
researcher and a better person every single day. I will always cherish
our numerous technical/non-technical conversations through which I
always found a way of balancing all aspects of life. Thank you Sir, for
imparting me the mantra - “Try to lead the wave, not to ride it.” I will
always lead my life that way.
A special thank you goes to Dr. Massimo Tornatore from Politecnico di
Torino, for all the help, feedback, and suggestions. Through enormous
discussions and brainstorming with him, it was possible to develop sev-
eral methods presented in this dissertation. His thoughtful and immac-
ulate insights have always shaped up my research. This dissertation
would not have been possible without his continued mentoring.
I would also like to thank my PhD committee members - Dr. Dipak
Ghosal and Dr. Xin Liu, from whom I got inspirations, insights, and
valuable suggestions on improving the quality of the dissertation. Dr.
Ghosal has always been an enormous source of knowledge and ideas.
His sharing of research experience defined many important aspects of
this dissertation. I have also greatly benefitted from the knowledge I
gathered from courses offered by both Dr. Ghosal and Dr. Liu. Dr. Liu
-x-
has motivated me further on improving my mathematical and analytical
abilities. I also take this opportunity to thank my two other qualifying
exam committee members - Dr. Matthew K. Farrens and Dr. Chen-Nee
Chuah, who have provided valuable feedback on my research.
I want to acknowledge a few things regarding Chapter 2 of this disser-
tation. This work has been done jointly by Yi Zhang, myself, Dr. Massimo
Tornatore, and Dr. Biswanath Mukherjee. In that work, I have made sig-
nificant contributions in the introductory section and core- and access-
network-related sections. I have included other parts of that work in
this dissertation to give a complete review on the energy efficiency of the
telecom optical networks.
I am indebted to National Science Foundation (NSF) (Grant CNS-0832-
176) for funding my research and Teknovus Inc. for donating optical ac-
cess equipment for experiments in Chapter 3. I would like to thank Dr.
Ezra Ip and Dr. Ting Wang of NEC Laboratories, America for helping us
to estimate the reaches of different line rates in Chapter 5.
I genuinely admire the opportunity of being a member of the Networks
Research Lab at UC Davis. I thank all my present and past labmates for
giving me such a pleasant and enthralling atmosphere to work. I would
specially like to thank Dr. Rajesh Roy, Avishek Nag, Dr. Suman Sarkar,
Dr. Lei Song, Dr. Huan Song, Dr. Marwan Batayneh, Dr. Dragos An-
drei, Dr. Vishwanath Ramamurthi, Dr. Cicek Cavdar, Dr. Joon-Ho Choi,
Dr. Ming Xia, Dr. Eiman Al-Otaibi, Dr. Davide Cuda, Dr. Ananya Das,
Dr. Marilet De Andrade Jardin, Yi Zhang, Menglin Liu, Abu (Sayeem)
Reaz, Lei Shi, Ferhat Dikbiyik, S. K. Chaitanya Vadrevu, Uttam Mandal,
Rui (Richard) Wang, Farhan Habib, Partha Bhaumik, Sai Gopal Thota,
Xiuzhong (Adam) Chen, and Shuqiang Zhang for their continued sup-
port. I also want to thank all the Department of Computer Science office
staff and faculty members for supporting various aspects of my graduate
-xi-
study at UC Davis.
I would also like to thank my friends and family members for pro-
viding me the cushion of support and encouragement throughout my
life. Finally, I extend my deepest gratitude to my parents, my brother,
my wife, and my daughter, knowing that nothing is enough to express
my sincerest feelings towards them. I am grateful to my parents, Dilip
Chowdhury and Anju Chowdhury, for helping me being a better human
being and supporting all my endeavors without question. Thank you
my dearest wife, Sanchita Dey, for encouraging and supporting me, and
walking all the steps with me. Thank you my precious little one, Mahika
- you have changed my life forever.
-xii-
Pulak ChowdhuryJune 2011
Computer Science
Energy-Efficient Next-Generation Networks (E2NGN)
Abstract
With increasing energy consumption of the Internet, it is now impera-
tive to design and develop energy-efficient next-generation network archi-
tectures and protocols. This dissertation presents novel and innovative
techniques and methods for developing energy-efficient next generation
telecom networks, both in core and access networking domains.
This dissertation begins with an introduction to energy efficiency in
telecom networks and its importance for sustainable network develop-
ment, along with the compilation of research contributions. Then, a
comprehensive literature review on the energy efficiency research in tele-
com optical networks is presented in Chapter 2. This survey works as
the outline of various research aspects in telecom networks’ energy effi-
ciency and provides a comparison of various energy-efficiency methods.
In Chapter 3, we present a working prototype for Wireless-Optical
Broadband Access Network (WOBAN), a next-generation, cost- and ener-
gy-efficient access network architecture. This prototype is flexible to
incorporate new technologies and protocols and facilitates researchers
to develop, build, and test their protocols for future broadband access
networks. We have experimented with several cutting-edge, media-rich
applications on top of our WOBAN prototype. This prototype also en-
ables researchers to examine the impact of energy-saving mechanisms
on WOBAN’s performance.
Next, we focus on developing energy-efficient network protocols and
architectures. We present models and algorithms to build an energy-
efficient WOBAN in Chapter 4. Future-proof access technologies should
xiii
be energy efficient. The techniques developed in Chapter 4 should enable
a “green” deployment of WOBAN.
In Chapter 5, we show design models for developing energy-efficient
Mixed-Line-Rate (MLR) optical networks. In an MLR network, a single
link can carry various line-rate signals. We examine and analyze differ-
ent energy-efficient MLR network architectures and compare the energy
efficiency of MLR networks with SLR networks (where a single link carries
same line-rate signals). Finally, concluding remarks and future research
directions are presented in Chapter 6.
In summary, this dissertation makes important contributions to the
networking-research knowledge base by presenting new methods, algo-
rithms, and models to design energy-efficient next-generation networks.
xiv
1
Chapter 1
Introduction
Information and Communication Technology (ICT) has provided us the
opportunities for transforming our society with environment-friendly tech-
nologies (e.g., online shopping, teleworking, remote communication, vir-
tualized office environment, smart buildings, etc.), thereby reducing hu-
man impact on nature. ICT enables other business sectors to visualize
and optimize their energy needs and green house gas (GHG) emissions to
make them more energy efficient. At the same time, the ICT (and the In-
ternet) is enhancing our lifestyle needs, increasing productivity, and sup-
porting economic developments across the world. Consequently, there is
an enormous growth of Internet traffic over the last several years, and it
will continue to grow at a faster rate in the upcoming years. More traffic
means more ICT equipment need to be installed, thereby increasing the
energy consumption of the ICT sector itself. Therefore, as the ICT sector
is growing rapidly, now is the time to ask another question: what impact
does pervasive ICT have on energy consumption and GHG emissions?
Energy consumption of ICT is increasing at a high rate since more
computers, networks, and communication equipment are being deployed
every year. It is estimated that ICT consumes around 8% of total elec-
tricity all over the world [1]. Telecom networks constitute a significant
part of ICT. With the growth of traffic volume in telecom networks, their
2
energy consumption is also increasing rapidly. If energy consumption
of the ICT sector continues to grow at an alarming rate, energy shortage
will represent an obstacle for future ICT and telecom network expansion.
Therefore, it is imperative to develop energy-efficient (“green”) network so-
lutions for sustainable ICT growth. Until recently, telecom researchers
mainly focused on designing networks with optimized resources (e.g.,
bandwidth, cost, etc.). With the increasing energy demand of the In-
ternet, it is now imperative to satisfy another design objective - energy
efficiency.
Energy-efficiency problems in telecom networks are being addressed
at different levels - from devices to network level to application level. In
this dissertation, our goal is to present novel telecom network architec-
tures and protocols which will help us to develop future energy-efficient
telecom networks. Telecom networks can be subdivided into three net-
work domains - core, metro, and access. Core networks geographically
cover nation/continent-wide distances, typically connect larger cities in
a country/continent, and have link lengths of few hundreds to few thou-
sands of kilometers. Metro networks cover a metropolitan region and
have link lengths of few tens to few hundreds of kilometers. Ideally, a
core network can connect several metro networks. The access network
extends the “last mile” of the telecom network hierarchy connecting the
end users to the network service providers. Optical network technologies
have been proven to be the front-runner among all the technologies in
all three network domains. High capacity, reliability, and low operating
costs have made optical technologies to become the frequently-chosen
physical infrastructure for future telecom networks. In this disserta-
tion, we address energy-efficiency issues in the core and access network
domains, with specific emphasis on optical technologies.
3
1.1 Energy Efficiency in Core NetworksCore network is the center piece of the telecom network hierarchy. In the
core, the main energy consumers are transmission and switching equip-
ment such as routers, OXCs (Optical Cross-Connects), amplifiers, and
transponders. Device vendors are always working on developing more
energy-efficient next-generation transmission and switching technolo-
gies. Unfortunately, operational core network architectures do not have
many energy-efficiency measures installed at the networking level (e.g.,
protocols, etc.) although core network energy consumption is increasing
at an alarming rate. Currently, many research efforts are focusing on
improving the energy efficiency of core networks. We can broadly clas-
sify the energy-efficiency research in networking into two subcategories:
energy-efficient network design and energy-efficient network operation.
In energy-efficient network design, researchers develop network archi-
tectures and protocols which will lead to improved network utilization,
thereby reducing energy cost per bit of transport. Designing energy-
efficient packet-forwarding and green routing protocols also falls in this
category. In energy-efficient network operation, measures such as selec-
tively turning off network elements and on-demand resource allocation
schemes can improve the energy efficiency of the network. Researchers
are investigating all these areas to develop sustainable core network ar-
chitectures, as well as metro networks.
1.2 Energy Efficiency in Access NetworksWhile legacy access technologies (such as Digital Subscriber Line (DSL)
and Cable Modem (CM)) show bandwidth constraints for the future, optical-
fiber-based technologies (e.g., fiber-to-the-home (FTTH), fiber-to-the-buil-
ding (FTTB), fiber-to-the-curb (FTTC), etc.) are showing promises to sup-
port high-bandwidth digital services. The next generation of access net-
4
works is therefore showing trends of deploying optical fiber all the way to
the customer premises. Recent technology developments have enabled
the network operators to deploy optical access networks, such as Passive
Optical Network (PON) [2] and improve the bandwidth capacity of the ac-
cess network. A recent study also suggested that PON is one of the most
energy-efficient access network solutions [3].
However, challenges exist in the deployment of an all-optical access
network. Cost of deployment and maintenance of a fiber-based optical
access network is very significant, especially for larger countries such
as the USA and especially for communities where the population den-
sity is not very high. On the other hand, wireless access technologies
support mobility and untethered access and provide ease of deployment
and cost effectiveness. Unfortunately, wireless access is constrained due
to limited bandwidth. Combining the complementary features of these
two technologies (optical and wireless) can potentially provide ubiqui-
tous (“anytime-anywhere") broadband access to satisfy future customer
demands. Therefore, a novel cross-domain network architecture – called
Wireless-Optical Broadband Access Network (WOBAN) – which is an opti-
mal combination of high-capacity optical backhaul and untethered wire-
less access, is proposed in the literature [4].
1.2.1 WOBAN and Energy Efficiency
WOBAN is emerging as a promising technology to provide economical and
scalable broadband Internet access. In this cross-domain network archi-
tecture, end users receive broadband services through a wireless mesh
front-end which is connected to the optical backhaul via gateway nodes.
WOBAN shows excellent promise for future access networks. WOBAN
saves on network deployment cost because fiber need not penetrate to
each end user, and it extends the reach of emerging optical access so-
lutions such as PON. How far the fiber should penetrate before wireless
5
front-end takes over is an interesting engineering optimization problem
[4]. Due to its high-capacity optical backhaul, the WOBAN’s transport
capacity is higher than the relatively low capacity of the wireless mesh
network.
WOBAN also exhibits excellent opportunities to improve the network
utilization, and hence the energy efficiency of access networks. WOBAN
takes advantage of the flexible wireless front-end and can reroute traffic
through alternate paths in case of failures such as a fiber cut, or other
failures. The flexibility provided by the wireless front-end of WOBAN
can also be efficiently exploited to enable energy savings in the optical
part. During low-load hours, we can selectively put some of the network
parts to sleep and reroute the traffic through the other parts of the net-
work, thanks to the flexible wireless front-end of WOBAN. Considerable
amount of energy savings can be obtained in WOBAN through intelligent
management techniques.
1.3 Research ContributionsThis dissertation presents four novel contributions in the area of energy
efficiency in telecom optical networks. Below, we briefly describe the
contributions.
1.3.1 Energy Efficiency in Telecom Optical Networks
In Chapter 2, we provide a comprehensive survey of the most relevant
research activities for minimizing energy consumption in telecom net-
works. We investigate the energy-minimization opportunities enabled by
optical technologies and classify the existing approaches over different
network domains, namely core, metro, and access networks. A section
is also devoted to describe energy-efficient solutions for some of today’s
important applications using optical network technology, e.g., grid com-
puting and data centers. We provide an overview of the ongoing stan-
6
dardization efforts in this area. This work presents a comprehensive and
timely survey on a growing field of research, as it covers most aspects of
energy consumption in optical telecom networks.
1.3.2 WOBAN Prototype and Research Challenges
Chapter 3 presents the procedures and issues related to developing a
WOBAN prototype. For successful integration of WOBAN in the opera-
tional networks, it is important to understand deployment issues, risks
associated with the deployment, and performance of WOBAN protocols in
practical scenarios. An experimental WOBAN prototype can reveal these
issues related to the federation of this access technology with the exist-
ing network infrastructure. This prototype will facilitate researchers to
pursue experimental networking research employing broadband access
using both optical and wireless access. In future, researchers can also
investigate energy-conservation mechanisms (as presented in Chapter 4)
and their impact on WOBAN performance using the prototype.
The WOBAN prototype serves as a testbed for various access network
protocols and data dissemination techniques; and it features (a) pro-
grammability - ability to create, modify and test protocols on the network,
(b) resource sharing - sharing of network resources among experiments,
and (c) slice-based experimentation - different experiments can be run
independently in different network partitions.
1.3.3 Building a Green WOBAN
In Chapter 4, we show how we can build a very-high-throughput “green”
hybrid wireless-optical broadband access network (WOBAN). We devise
novel energy-saving techniques for WOBAN to improve its energy effi-
ciency and network utilization. We present a Mixed Integer Linear Pro-
gram (MILP) model which acts as a benchmark for evaluating our tech-
niques. We analyze the impact of energy-aware design and protocols
7
on the performance of WOBAN over dynamic traffic profiles. Illustrative
numerical examples show that, with suitable design parameters, we can
efficiently reduce energy consumption in WOBAN without significantly
impacting the network performance.
1.3.4 Energy-Efficient Mixed-Line-Rate Network Design
While optical technologies have shown significant promise in improv-
ing the energy efficiency of the network infrastructures, there are still
enormous efforts needed to increase the energy efficiency of optical net-
works. Future optical backbone networks will be heterogeneous in na-
ture where a single link may carry various line-rate signals. This Mixed-
Line-Rate (MLR) network architecture will cost-effectively satisfy hetero-
geneous traffic demands. In Chapter 5, we present mathematical mod-
els to design energy- and cost-efficient MLR optical networks. We also
perform a comparative study of the energy efficiency of MLR and single-
line-rate (SLR) networks (where all the links have same line rates). The
results show that MLR networks are more energy efficient than SLR net-
works.
1.4 OrganizationThe rest of the dissertation is organized as follows:
Chapter 2 presents a detailed survey on the energy efficiency of tele-
com optical networks. In this survey, our goal is to provide a compre-
hensive reference for the growing base of researchers who will work on
energy efficiency of telecom networks in the upcoming years. This work
has appeared in IEEE Communications Survey and Tutorials, October
2010 [5].
In Chapter 3, we present the architecture and functional character-
istics of a WOBAN prototype built in the Networks Lab at UC Davis. We
cite some research challenges on hybrid networks based on our exper-
8
imental observations. This work has appeared in IEEE Network, May
2009 [6].
With the increasing energy consumption of the Internet, it is now im-
perative to design and develop energy-efficient network architectures and
protocols. In Chapter 4, we provide algorithms and methods for deploy-
ing an energy-efficient WOBAN. This work has appeared in IEEE/OSA
Journal of Lightwave Technology (JLT), August 2010 [7], after presenta-
tion at IEEE GLOBECOM Conference, December 2009 [8].
Chapter 5 develops models to design energy-efficient MLR optical net-
works. It also examines the energy efficiency of MLR networks com-
pared to Single-Line-Rate (SLR) networks. This work is under review in
IEEE/OSA Journal of Lightwave Technology (JLT), after presentation at
the Optical Fiber Communications (OFC) Conference, March 2010 [9].
Finally, Chapter 6 concludes the dissertation.
9
Chapter 2
Energy Efficiency in TelecomOptical Networks
2.1 IntroductionEnergy conservation is gaining increasing interest in our society in re-
cent years. There is growing consensus on the necessity to put energy
conservation at the top of the research agenda, as one of the most com-
pelling and critical current research issues. Today, traditional energy
resources, such as hydrocarbon energy, provide most of the energy de-
mand, e.g., 85 percent of primary energy of USA’s electricity [10], but
this kind of energy is not renewable, and it is expected to be finally used
up in the not-too-distant future. Besides, the combustion of hydrocar-
bon materials releases large amounts of Green House Gases (GHG), a
major cause of Global Warming.
Two research directions are being explored to address this situation.
First, renewable energy is being harnessed to replace traditional hy-
drocarbon energy. This not only gives the opportunity to reduce the
carbon footprint, but also it paves the road towards a sustainable and
environment-friendly societal development [11]. Second, energy-conserva-
tion approaches are being investigated in many science and technology
areas - low-energy equipment and components are being developed, not
10
only to decrease the energy cost, but also to help to save our environ-
ment. In almost all scientific disciplines where technological develop-
ment may allow to reduce the amount of energy needed to support hu-
man activities, research efforts are ongoing to devise new solutions for
energy conservation.
Information and Communication Technology (ICT) is one of the most
promising areas for pursuing energy conservation. ICT is widely used in
most aspects of our society and has traditionally had an environment-
friendly image. This good reputation comes mostly from the fact that
worldwide telecom networks have transformed our society and provided
practical means to reduce the human impact on nature (consider, for ex-
ample, telecom applications for telework, videoconference, e-commerce,
and their reduced impact on human movements). There is however a
downside of ICT. The ubiquitousness of ICT in daily life (both private and
professional) brings another issue - the energy consumption of comput-
ers and network equipment is becoming a significant part of the global
energy consumption [12], [13], [14].
As the coverage of ICT is spreading rapidly worldwide, the energy con-
sumption of ICT is also increasing fast, since more equipment and com-
ponents for networks and communications are being deployed annually.
From the data of 2009, ICT consumes about 8% of the total electricity
all over the world [1]. Telecom networks, which represent a significant
part of the ICT, are penetrating further into our daily lives. The traffic
volume of broadband telecom networks is increasing rapidly and so is its
energy consumption. Figure 2.1 reports a prediction of the energy con-
sumption growth (by percentage) of telecom networks in the coming years
[15], [16]. Considering both the growing energy price (expected with the
decline of cheap availability of fossil fuels) and the increasing concern on
the Green House effect which is being translated in government policies,
11
the energy consumption of ICT is already raising questions, and it is im-
perative to develop energy-efficient telecom solutions. We need to design
new networking paradigms so that ICT will maintain the same level of
functionality while consuming a lower amount of energy in future [12],
[17].
Figure 2.1. Energy consumption forecast of telecom networks [15], [16].
Among the various network technologies, in this work, we mainly fo-
cus on energy efficiency of optical networking technologies. Optical tech-
nologies are widely used in telecom networks, and currently they con-
stitute the basic physical network infrastructure in most parts of the
world, thanks to their high speed, large capacity, and other attractive
properties [18]. Optical networking technologies have also improved sig-
nificantly in the recent decade. Different characteristics of optical net-
works have been investigated and many approaches have been proposed
to improve the performance of optical networks. For instance, routing,
wavelength assignment, and traffic grooming strategies have been pro-
posed to make the optical network more cost-efficient [19]. Survivability
of optical networks has also been thoroughly investigated because a fail-
ure of an optical link or node can cause a significant loss due to the large
bandwidth of an optical communication channel [20].
12
Nevertheless, the energy-efficient optical network is a new concept,
which is being investigated in recent years. More research groups are
starting to focus on it since energy-efficient optical networks will con-
tribute to save the energy consumed by ICT, and further reduce the en-
ergy consumption of our society and protect our environment.
Minimizing energy consumption of optical networks can be generically
addressed at four levels: component, transmission, network, and appli-
cation. At the component level, highly-integrated all-optical processing
components such as optical buffers, switching fabrics, and wavelength
converters are being developed, which will significantly reduce energy
consumption [21], [22]. Optical Switching Fabric (OSF) is more energy-
efficient than electronic backplanes and interconnects [23], [24]. At the
transmission level, low-attenuation and low-dispersion fibers, energy-
efficient optical transmitters and receivers, which improve the energy ef-
ficiency of transmission, are also being introduced [25]. Energy-efficient
resource allocation mechanisms, green routing, long-reach optical ac-
cess networks [26], etc. are being investigated at the network level to
reduce energy consumption of optical networks. At the application level,
mechanisms for energy-efficient network connectivity such as “Proxying”
[27] and green approaches for cloud computing [28] are being proposed
to reduce the energy consumption.
Here, our objective is to mainly survey the energy-saving approaches
at the network level. Typically, a telecom network can be subdivided
into three domains: core, metro, and access. Optical technologies play
a relevant role in each of these domains, and we survey the research
efforts to improve the energy efficiency of optical network solutions in all
three domains.
As shown in Fig. 2.2, the core network is the central part of the tele-
com hierarchy, and it provides nationwide or global coverage. Links in
13
the core network span long distances – a link (employing optical fibers)
could be a few hundreds to a few thousands of kilometers in length, e.g.,
links providing connections between the main cities of the Unites States.
Typically, core networks rely on mesh topologies that provide increased
protection flexibility and efficient utilization of network resources. The
metro network typically spans a metropolitan region, covering distances
of a few tens to a few hundreds of kilometers and is dominantly based
on a deep-rooted legacy of SONET/SDH optical ring networks. The ac-
cess network connects the end users to their immediate service provider.
The access network enables end users (businesses and residential cus-
tomers) to connect to the rest of the network infrastructure, and it spans
a distance of a few kilometers. Optical access networks are usually based
on tree-like topologies.
Figure 2.2. Telecom network hierarchy.
In this work, energy consumption data and energy-conservation ap-
proaches are surveyed in all three network domains. We also review some
relevant energy-saving approaches in the application layer and energy-
14
efficient architectures in data centers because these domains: (i) involve
network elements that consume significant energy in a telecom network,
and (ii) they largely involve optical networking technologies.
A comprehensive survey on new solutions for energy-efficient opti-
cal networks is a very timely and useful contribution since researchers
working on energy-efficient optical networks may benefit from having a
handy collection of basic information on the energy consumption of the
various components of an optical network as their background of re-
search, and also a comprehensive classification with comments on cur-
rent efforts and approaches can inspire researchers to have new ideas
on energy-saving research. Our survey includes these two aspects and
anticipates possible future research areas. Also, note that various in-
ternational standardization organizations, such as ITU (International
Telecommunication Union), IEEE (Institute of Electrical and Electronics
Engineers), and others, are currently working on developing new stan-
dards to strengthen research on this topic [29]. In this work, we also
include a summary of these standardization efforts.
The rest of the chapter is organized as follows. Section 2.2 classi-
fies the network domains on which optical technologies are employed,
and provides energy consumption data for the optical components and
systems used in various network domains. Section 2.3 summarizes the
standardization efforts for energy-efficient telecom network design. Sec-
tion 2.4 provides an overview of techniques and architectures for energy-
consumption minimization in core networks, while Section 2.5 provides
the corresponding treatment for optical metro and access networks. Sec-
tion 2.6 describes some recent approaches on how optical networking
technologies can be employed to increase the energy efficiency in data
centers and in the application layer. Finally, Section 2.7 concludes the
chapter.
15
2.2 Network DomainsTelecom networks can be divided into three network domains: core, metro,
and access (Fig. 2.2). Optical technologies have been applied in all these
network domains in order to support higher transmission rates and more
cost-effective data transfer. In this section, we describe the three net-
work domains and introduce the most important network elements of
each domain. For each of these network elements, we also provide rep-
resentative data and references regarding their energy consumption.
2.2.1 Core Network
By core network, we usually refer to the backbone infrastructure of a tele-
com network, which interconnects large cities (as network nodes), and
spans nationwide, continental, and even intercontinental distances. The
core network is typically based on a mesh interconnection pattern and
carries huge amounts of traffic collected through the peripheral areas
of the network. So, it needs to be equipped with appropriate interfaces
towards metro and access networks which are in charge to collect and
distribute traffic, so that users separated by long distances can commu-
nicate with one another through the core (backbone) network.
In the core network, optical technologies are widely used to support
the basic physical infrastructure and achieve high speed, high capacity,
scalability, etc. To intelligently control and manage the optical network,
several high-level management equipment and technologies have been
developed. For example, network architectures based on IP (Internet
Protocol) over SONET / SDH (Synchronous Digital Hierarchy), IP over
WDM (Wavelength-Division Multiplexing), or IP over SONET/SDH over
WDM have been deployed over the past two decades [30], [31]. As core
networks exhibit multi-layer network architectures, energy consumption
of the core network should be considered at both of the network layers,
16
Figure 2.3. Core network.
i.e., the optical layer and the electronic layer.
Let us consider an IP-over-WDM network as an example, as shown in
Fig. 2.3 - energy consumption of its network components can be found
in the switching (routing) level and also in the transmission level. In the
switching (routing) level, the main energy consumers are Digital Cross-
Connects (DXC) and IP routers for switching electric signals at the elec-
tronic layer, while Optical Cross-Connects (OXC) are used to switch op-
tical signals in fibers at the optical layer. In the transmission systems,
WDM is a technology which multiplexes multiple optical carrier signals
on a single optical fiber by using different wavelengths of laser light to
carry different signals. As shown in Fig. 2.3, a WDM transport system
[32] uses a multiplexer at the transmitter to join the signals together,
and a demultiplexer at the receiver to split them apart. Transponders
are used for transmitting and receiving signals. The booster is a power
amplifier which can compensate the power loss caused by the multi-
17
plexer. The pre-amplifier is used to amplify the power of optical signals
so as to increase the sensitivity of the receiver. All these components
of a WDM transmission system consume energy. Erbium-Doped Fiber
Amplifiers (EDFAs), which are used for amplifying optical signals in the
optical fiber, also consume energy: the energy consumption of an opti-
cal amplifier and how to measure it may depend on the way the optical
amplifier is operated [33].
Next, we provide some typical data on energy consumption of the most
important network components in core networks. Table 2.1 shows the
energy consumption data of these components. The power values re-
ported in Table 2.1 are associated with the maximum load that the cor-
responding equipment can serve, except for all-optical equipment where,
due to the transparency of the system to the bit rate, power value at a
specified aggregate rate is difficult to calculate. Nonetheless, the above-
mentioned property of transparency makes optical equipment more scal-
able (to increase capacity) than electronic equipment. By analyzing these
data, it clearly emerges that energy consumed by the electronic layer is
much larger than that of the optical layer. In other words, optical switch-
ing is more energy-efficient than electronic switching which is one of the
basic ideas for energy-efficient network design by exploiting optical tech-
nology.
2.2.2 Metro Network
The metro network is the part of a telecom network that typically covers
metropolitan regions. It connects equipment for aggregation of residen-
tial subscribers’ traffic (e.g., it provides interfaces to dispersed access
network, such as various flavors of Digital Subscriber Line (xDSL) and
Fiber-to-the-Home or Fiber-to-the-x (FTTx)), and it provides direct con-
nections to the core network for Internet connectivity. Different network-
ing technologies have been deployed in different metro areas across the
18
Table 2.1. Typical power values of different components.
Network
Domain
Component Capacity Energy Con-
sumption
Core
Network
Core Router (Cisco CRS-1 Multi-
shelf System)
92
Tbps
1020 kW [34]
Optoelectronic Switch (Alcatel-
Lucent 1675 Lambda Unite Multi-
Service Switch)
1.2
Tbps
2.5 kW [35]
Optical Cross-Connect (MRV Op-
tical Cross-Connect)
N/A 228 W [36]
WDM Transport System (Ciena
CoreStream Agility Optical Trans-
port System)
3.2
Tbps
10.8 kW [32]
WDM transponder (Alcatel-Lucent
WaveStar OLS WDM Transponder)
40
Gbps
73 W [37]
EDFA (Cisco ONS 15501 EDFA) N/A 8 W [37]
Metro
Network
Edge Router (Cisco 12816 Edge
Router)
160
Gbps
4.21 kW [38],
[39]
SONET ADM (Ciena CN 3600 Intel-
ligent Optical Multiservice Switch)
95
Gbps
1.2 kW [40]
OADM (Ciena Select OADM) N/A 450 W [41]
Network Gateway (Cisco 10008
Router)
8 Gbps 1.1 kW [39]
Ethernet Switch (Cisco Catalyst
6513 Switch)
720
Gbps
3.21 kW [34],
[39]
Access
Network
OLT (NEC CM7700S OLT) 1 Gbps 100 W [3]
ONU (Wave7 ONT-E1000i ONU) 1 Gbps 5 W [3]
19
Figure 2.4. Metro and access networks.
world. As shown in Fig 2.4, SONET (Synchronous Optical Networking),
Optical WDM ring, and Metro Ethernet are three dominant technologies
in metro networks. As an example, Metro Ethernet is a commonly-used
metro network infrastructure which is based on the Ethernet standard
[42] - edge routers, broadband network gateways, and Ethernet switches
are its basic components. Energy consumption data of some Metro Eth-
ernet equipment are shown in Table 2.1.
Metro WDM ring networks have also been proposed to take the ad-
vantages of optical technology, such as higher speed and more scala-
bility [43]. In metro WDM ring networks, energy consumption comes
mainly from OADMs (Optical Add-Drop Multiplexers) which are used to
add and drop optical signals. SONET ring architectures are also widely
deployed in metro networks, which can aggregate low-bit-rate traffic of
metro networks to high-bandwidth pipes of core networks [18]. SONET
ADM (Add-Drop Multiplexer) is used to add and drop network traffic.
Energy consumption of a SONET ADM is shown in Table 2.1.
20
2.2.3 Access Network
The access network is the “last mile” of a telecom network connecting the
telecom CO (Central Office) with end users. Access network comprises
the larger part of the telecom network. It is also a major consumer of
energy due to the presence of a huge number of active elements [15].
There are several access technologies proposed and deployed in the
market such as xDSL (Digital Subscriber Line), CM (Cable Modem), Wire-
less and Cellular networks, FTTx, WOBAN (Wireless-Optical Broadband
Access Network), etc. These technologies can be broadly classified into
two categories – (a) wired (such as xDSL, CM, FTTx, etc.) and (b) wireless.
The enhanced copper or xDSL systems cover various technologies
such as ADSL (asymmetric DSL), VDSL (very-high-speed DSL), and HDSL
(high-bit-rate DSL). xDSL technologies use existing PSTN (Public-Switch-
ed Telephone Network) infrastructure to provide Internet service. Cable
modem technology uses co-axial cable to provide Internet service along
with digital TV. FTTx has different underlying technologies, such as di-
rect fiber, shared fiber, and the most dominant one - PON (Passive Optical
Network).
PON is the leading choice for fiber access network deployment be-
cause it has only passive elements in the fiber plant (see Fig. 2.4). Ta-
ble 2.1 reports energy consumption data for the two main network el-
ements in a PON architecture: OLT (Optical Line Terminal), located at
the CO, and ONU (Optical Network Unit), located at (or close to) the end
customer. Wireless access technologies include WiFi (Wireless Fidelity),
WiMAX (Worldwide Interoperability for Microwave Access), and Cellular
data service (such as LTE (Long Term Evolution), etc.). WOBAN is a novel
access architecture which consists of a wireless network at the front-end
supported by an optical backhaul, and can provide high-bandwidth ser-
vice.
21
2.3 Standardization EffortsThe importance of energy efficiency in networking has also been acknowl-
edged by a number of new workgroups in international standards organi-
zations. Several of them, such as ITU (International Telecommunication
Union), IEEE (Institute of Electrical and Electronics Engineers), ETSI
(European Telecommunication Standard Institute), TIA (Telecommuni-
cation Industry Association), ATIS (Alliance for Telecommunications In-
dustry Solutions), ECR (Energy Consumption Rating) Initiative, TEEER
(Energy Efficiency Requirements for Telecommunications Equipment),
etc. are working on new standards for energy-efficient networks [29].
They are developing novel concepts for green networking and their ac-
tivity can provide guidance to researchers on the practicality of their
research.
As part of a major initiative on Green Networks, ITU is organizing Sym-
posia on ICT and Climate Change [44]. These symposia bring together
key specialists in the field: from top decision makers to engineers, de-
signers, planners, government officials, regulators, standards experts,
and others. Topics presented and discussed include the adaptation and
mitigation of the effects of climate change on the ICT sector and on other
sectors, “green” ICT policy frameworks, and the use of ICT in climate
change science and in emergency situations. The ITU Telecommunica-
tion Standardization Sector has also announced the establishment of
SG (Study Group)-15 on energy-conservation techniques. The technolo-
gies considered in the list include optical transport networks and access
network technologies such DSL and PON. Together, these technologies
represent a significant consumption of energy worldwide.
IEEE developed a standard on Energy-Efficient Ethernet - IEEE P802.-
3az [45]. Its objectives are (i) to define a mechanism to reduce power
consumption during periods of low link utilization for the PHYs (Physi-
22
cal layer protocol), (ii) to define a protocol to coordinate transitions to or
from a lower level of power consumption which do not change the link
status or drop frames, and (iii) to define a 10 megabit PHY with a re-
duced transmit amplitude requirement so that power consumption can
be decreased. IEEE ratified the final standard on October 2010 [45].
ETSI Green Agenda is one of ETSI’s major strategic topics [46]. This
effort will implement the ISO 14001:2004 and 14004:2004 standards
which are the Environmental Management Standards. In addition, ETSI
Green Agenda includes Environmental Engineering, which consists of
(i) “DTR/EE-00002” Work Item: reduction of energy consumption in
telecommunications equipment and related infrastructure; (ii) “DTR/EE-
00004” Work Item: use of alternative energy sources in telecommunica-
tion installations; (iii) “DTS/EE-00005” Work Item: energy consump-
tion in Broadband Telecom Network Equipment; (iv) “DTS/EE-00006”
Work Item: environmental consideration for equipment installed in out-
door location; and (v) “DTS/EE-00007” Work Item: energy efficiency of
wireless access network equipment. In addition, ETSI ATTM (Access,
Terminals, Transmission, and Multiplexing) “DTR/ATTM-06002” Work
Item, which is about power optimization of xDSL transceivers, is under
standardization. In the DTS/EE-00005 Work Item, which is the most
closely related to the topic of this work, ETSI leads the effort to define
energy consumption targets and measurement methods for both wired
and wireless broadband-telecom-network equipment. In the first phase,
DSL, ISDN (Integrated Services Digital Network), etc. have been consid-
ered. In the second phase, energy consumption for WiMAX, PLC (Power
Line Communication) will be investigated [46].
TIA started a “Green Initiative” in 2008, called EIATRACK [47]. It of-
fers companies a way to strategize their future growth and environmental-
ly-conscious initiatives in new markets. Its key product-compliance is-
23
sues are about Take-back, Batteries, Restricted Substances, Design for
Environment, and Packaging. More than 1,500 pieces of legislation are
tracked, from proposal through implementation, which cover all major
regions of Europe, Asia Pacific, North America, and South America. It
contains accurate, up-to-date content provided by a wide range of inter-
national legal and technical subject-matter experts, and EEE (Electrical
and Electronic Equipment) and RoHS (Restriction of the use of certain
Hazardous Substances) experts in Europe and other jurisdictions.
ATIS has set up a committee named NIPP (Network Interface, Power,
and Protection Committee), which is working on developing standards
and technical reports covering Network Interfaces, Power, Electrical, and
Physical Protection [48]. The “Green” activities of the NIPP committee
are focused on: (i) producing standards that may be used by Service
Providers to assess the true energy needs of telecom equipment, (ii) RoHS
in electronic equipment, and (iii) investigating methods to reduce the
power consumption of DSL modems at both network and customer ends
of the line [29]. The NIPP has also established the TEE (Telecommu-
nications Energy Efficiency) subcommittee which develops and recom-
mends standards and technical reports related to the energy efficiency of
telecommunication equipment. They are making efforts to define energy-
efficiency metrics, measurement techniques, as well as new technologies
and operational practices for telecommunications components, systems,
and facilities [49]. In summary, like the standardization organizations
listed above, ATIS is also focusing on “Green” technologies at both the
physical and the network layers.
The concept of ECR (Energy Consumption Rating) has also been ini-
tiated recently. Since governments and corporations around the world
are tightening energy consumption and carbon emission budgets, tele-
com equipment manufacturers are claiming to develop new and energy-
24
efficient equipment. Verifiable data is needed to support these “green”
marketing claims. ECR is defined to measure the energy efficiency of
network equipment which is expressed in Watts/Gbps. As a primary
metric, ECR is expressed to measure the ratio of power consumption
and transmission bandwidth. New criteria are also used to define the
practical aspects of energy efficiency for the networking industry [50].
TEEER Metric Quantification (Energy-Efficiency Requirements for Tel-
ecommunications Equipment) has been achieved from the Verizon energy-
efficiency initiative, VZ.TPR.9205. The purpose of this program is to set
Verizon technical purchasing requirements and to foster the development
of energy-efficient telecom equipment, thereby reducing GHG emissions.
TEEER is defined as an average rating of the power consumption of an
equipment at multiple utilization levels. TEEER metric applies to all new
equipment purchased by Verizon after January 1, 2009 [51].
2.4 Core NetworkIn core networks, energy is mostly consumed in network transmission
and switching equipment such as routers, OXCs (Optical Cross-Connects),
EDFAs, and transponders. Based on the data of Section 2.2, the amount
of energy consumed by core networks is huge. However, current network
architectures and operation schemes generally do not pay much atten-
tion to energy efficiency. Therefore, many recent research efforts focus
on energy-efficient core network. The approaches to reduce energy con-
sumption in core networks can be divided into four categories: (i) selec-
tively turning off network elements, (ii) energy-efficient network design,
(iii) energy-efficient IP packet forwarding, and (iv) green routing.
2.4.1 Selectively Turning Off Network Elements
A major approach to save energy in the core network consists of selec-
tively switching off idle network elements when traffic load decreases
25
(e.g., at night), while still maintaining the vital functions of the network
in order to support the residual traffic. If we consider a representation
of the network hierarchy as in Fig. 2.2, we can see that there is of-
ten enough redundancy in the network so that some of the nodes can
be completely turned off when they are not used as source or destina-
tion of traffic, and they are not essential also as transfer nodes. In this
context, a node can be turned off (i) only when it is totally unused, (ii)
when the traffic goes below a given threshold, leaving the responsibility
to reroute the residual traffic to upper layers, and (iii) after proactively
rerouting the traffic along other routes, in order to avoid traffic disrup-
tions. These three approaches involve a wide range of burdens as far
as control, management, and operation of the network are concerned.
While the first approach requires no or minimal additional network con-
trol and the second only requires to gather congestion information, the
third approach can be applied only in a network that has some form of
automatic provisioning and/or reprovisioning in place.
In a similar manner, links can be switched off when there is no traffic
on them, or when traffic goes below a given threshold, or when it is pos-
sible to re-route the traffic flowing along them. Unfortunately, most of
the elements in a core network can not be just shut down without affect-
ing the performance of the network. Shutting down an intermediate core
node may cause the connection to be rerouted over a longer route, which
may sometimes not be acceptable due to various reasons: congestion,
extra delay, etc. So, the possibility of turning off nodes or links has to
be carefully evaluated under connectivity and QoS (Quality-of-Service)
constraints.
This problem has been modeled in [52] over a specific case study net-
work - in order to maximize energy saving, one has to identify the maxi-
mum number of idle nodes and links while still supporting the ongoing
26
traffic. This problem has been proven to be a NP-hard problem and
can be formulated as a MILP (Mixed Integer Linear Program). Since the
problem is computationally intractable, heuristics have been proposed
in [53]. Moreover, traffic load varies at different hours of the day. As-
suming that traffic demand at off-peak time is up to 60% lower than that
at peak time, it is possible to reduce the percentage of powered nodes to
17% and links to 55% in the off-peak hours by switching off idle nodes
and links, while ensuring that the resource utilization is still within a
given threshold [54]. In [55] and [56], the authors discuss the relation-
ship between network robustness, performance, and Internet power con-
sumption based on data collected from Internet sources.
In [57], the authors deduce energy-efficiency limit of adaptive net-
works. They develop several traffic models based on real traffic observa-
tions. If networks can follow these traffic models during resource alloca-
tion where resources will be allocated according to the traffic demands,
energy efficiency of such networks can improve significantly from the
current mode of operation in networks where resources are always on ir-
respective of the traffic demands. In [58], a scheme is proposed to shut
down idle line cards (and the corresponding optical circuit or lightpath)
when the traffic load is low. In this scheme, the physical topology is not
changed and energy is saved by only changing the virtual connectivity.
Similarly, in [59], the authors have also proposed a scheme to save en-
ergy by shutting down idle line cards, and also chassis, of IP routers
in IP-over-WDM networks when the traffic load is low. In addition, this
scheme minimizes the potential traffic interruption when the line cards
and chassis are shut down.
2.4.2 Energy-Efficient Network Design
Another possible way to achieve energy efficiency is to devise energy-
efficient architectures during the network-design stage. For example,
27
in [37], the authors consider a design approach for an IP-over-WDM net-
work where the energy consumption of IP routers, EDFAs, and transpon-
ders is jointly minimized. The results show that different schemes of
traffic grooming have a significant impact on energy-efficient design [37].
In this work, heuristics have also been proposed to minimize the en-
ergy consumption of network equipment. The authors considered two
possible ways to implement IP-over-WDM networks, i.e., lightpath non-
bypass and bypass. Under lightpath non-bypass, all the lightpaths in-
cident to a node must be terminated, i.e., all the data carried by the
lightpaths is processed and forwarded by IP routers. But the lightpath
bypass approach allows IP traffic, whose destination is not the interme-
diate node, to directly bypass the intermediate router via a cut-through
lightpath. Results show that lightpath bypass can save more energy than
non-bypass, leading to the conclusion that the number of IP routers can
be decreased while using the lightpath-bypass scheme in designing an
energy-efficient core network. Besides, the authors also estimated the
energy consumption of routers, EDFAs, and transponders separately. It
is shown that the total energy consumption of routers is much more than
that of EDFAs and transponders in IP-over-WDM networks.
Line cards and chassis of core routers consume considerably higher
amount of energy in core networks. Different line card/chassis configu-
rations, i.e., different fill levels of the chassis, result in different energy
consumption. The higher the fill level is, the more energy-efficient the
network will be [60]. This is because even an empty chassis without
line cards consumes a large amount of energy. Therefore, a chassis with
higher fill level has lower energy consumption per transferred bit than
the ones with lower fill levels. Besides, even if two chassis have the same
throughput, the chassis which supports higher-speed line cards tends to
consume less energy (per bit) than the one which supports lower-speed
28
line cards [61]. Therefore, energy-efficient line card/chassis reconfigu-
ration can be a novel way to reduce energy consumption.
Future optical backbone networks will be required to support MLR
(Mixed Line Rates) (e.g., 10/40/100G) over its links (Chapter 5). In Chap-
ter 5, we present mathematical models to determine the energy efficiency
of MLR optical networks. We consider three different MLR network ar-
chitectures. We compare the energy consumption of both MLR and SLR
(Single Line Rate) networks using the models. The results indicate that
a MLR network performs better than the SLR networks by reducing the
networkwide energy consumption.
2.4.3 Energy-Efficient IP Packet Forwarding
Energy-aware packet forwarding has been proposed to lower energy con-
sumption at the IP layer. In [61], the authors show that the size of IP
packets impacts the energy consumption of routers. For a constant-bit-
rate traffic scenario, the smaller the IP packets the routers transfer, the
more energy they consume. Thus, new IP packet forwarding schemes
can be designed to be energy-efficient. The size of IP packets can be op-
timized to save energy when they are being forwarded through routers.
However, a tradeoff exists between packet switching delay and energy-
efficient IP packet forwarding.
29
Tabl
e2.
2:C
ompa
riso
nof
gree
ning
effor
tsin
core
netw
orks
.
Pape
rA
lgor
ithm
Ene
rgy
Cos
tR
etro
fit
Deg
ree
ofE
nerg
ySa
ving
sE
xtra
Sign
alli
ng
and
Con
trol
App
roac
h
L.C
hiar
avig
lioet
al.
[52]
,[53
],[5
4]
MIL
P&
Heu
rist
ics
[52]
,[5
3],
Heu
rist
ics
[54]
Min
imiz
edC
ompl
iant
Hig
h(s
hutt
ing
dow
nid
leno
des)
Yes
Sele
ctiv
ely
turn
ing
offne
t-
wor
kel
emen
ts
F.Id
ziko
wsk
iet
al.
[58]
MIL
PM
inim
ized
Com
plia
ntH
igh
(shu
ttin
gdo
wn
idle
line
card
sof
rou
ters
)
Yes
Sele
ctiv
ely
turn
ing
offne
t-
wor
kel
emen
ts
Y.Zh
ang
etal
.[5
9]M
ILP
Min
imiz
edC
ompl
iant
Hig
h(s
hutt
ing
dow
nid
lelin
e
card
san
dC
hass
isof
rou
ters
)
Yes
Sele
ctiv
ely
turn
ing
offne
t-
wor
kel
emen
ts
C.L
ange
etal
.[5
7]N
/AN
on-m
inim
ized
New
Hig
h(a
dapt
ive
netw
orks
)Ye
sSe
lect
ivel
ytu
rnin
goff
net-
wor
kel
emen
ts
G.S
hen
etal
.[3
7]M
ILP
&
Heu
rist
ics
Min
imiz
edC
ompl
iant
Hig
h(e
nerg
ym
inim
izin
gin
two
laye
rs)
No
Ene
rgy-
effici
ent
netw
ork
desi
gn
P.C
how
dhu
ryet
al.
[9]
MIL
PM
inim
ized
New
Hig
h(M
ixed
-Lin
e-R
ate
netw
orks
)N
oE
nerg
y-effi
cien
tne
twor
k
desi
gn
L.C
eupp
ens
[60]
N/A
Non
-min
imiz
edC
ompl
iant
Low
(cha
ssis
reco
nfigu
rati
on)
No
Ene
rgy-
effici
ent
netw
ork
desi
gn
M.B
aldi
etal
.[6
2]N
/AN
on-m
inim
ized
New
Med
ium
(pip
elin
efo
rwar
ding
)Ye
sE
nerg
y-effi
cien
tIP
pack
et
forw
ardi
ng
J.C
haba
rek
etal
.[6
1]M
ILP
Min
imiz
edC
ompl
iant
Med
ium
(ene
rgy
min
imiz
ing
inIP
laye
r)
Yes
Ene
rgy-
effici
ent
IPpa
cket
forw
ardi
ng&
Gre
enro
uti
ng
S.Fi
guer
ola
etal
.[6
3]N
/AN
on-m
inim
ized
New
Hig
h(r
enew
able
ener
gyu
tiliz
a-
tion
)
Yes
Gre
enro
uti
ng
Con
tinu
edon
next
page
...
30
Tab
le2.
2–
cont
inue
dfr
ompr
evio
uspa
ge
Pape
rA
lgor
ithm
Ene
rgy
Cos
tR
etro
fit
Deg
ree
ofE
nerg
ySa
ving
sE
xtra
Sign
alli
ng
and
Con
trol
App
roac
h
B.S
t.A
rnau
d[6
4]N
/AN
on-m
inim
ized
New
Hig
h(r
enew
able
ener
gyu
tiliz
a-
tion
)
Yes
Gre
enro
uti
ng
E.Y
etgi
ner
etal
.[6
5]M
ILP
Min
imiz
edC
ompl
iant
Med
ium
(tra
ffic
groo
min
g)N
oG
reen
rou
ting
M.X
iaet
al.
[66]
[67]
Heu
rist
ics
Non
-min
imiz
edC
ompl
iant
Med
ium
(tra
ffic
groo
min
g)N
oG
reen
rou
ting
B.P
uyp
eet
al.
[68]
Heu
rist
ics
Non
-min
imiz
edC
ompl
iant
Med
ium
(tra
ffic
groo
min
g)N
oG
reen
rou
ting
S.H
uan
get
al.
[69]
MIL
P&
Heu
rist
ics
Min
imiz
edC
ompl
iant
Med
ium
(tra
ffic
groo
min
g)N
oG
reen
rou
ting
Y.W
uet
al.
[70]
MIL
P&
Heu
rist
ics
Min
imiz
edC
ompl
iant
Med
ium
(rou
ting
and
wav
elen
gth
assi
gnm
ent)
No
Gre
enro
uti
ng
M.
Has
anet
al.
[71]
,
[72]
Heu
rist
ics
Non
-min
imiz
edC
ompl
iant
Med
ium
(tra
ffic
groo
min
g)N
oG
reen
rou
ting
31
Another approach for energy-efficient IP packet forwarding is pipeline
forwarding [73]. It is a “time-based” IP packet-switching scheme (also re-
ferred to as Time-Driven Switching), and it enables to extend the energy-
efficient time-based IP packet switching all the way to the edges of the
network. Based on pipeline forwarding, a network architecture which in-
cludes two independent, tightly-integrated, parallel subnetworks is pro-
posed in [62]. The two subnetworks are the current Internet and “super-
highways” where pipeline forwarding of IP packets is deployed (Fig. 2.5).
Besides carrying typical traffic, such as mail, low-priority web browsing,
and file transfers, asynchronous IP routers are used to transport the sig-
naling required to set up synchronous virtual pipes in the pipeline for-
warding parallel network which carries traffic requiring a deterministic
service, such as phone calls, video on demand, video conferencing, and
distributed gaming. Large bandwidth is required by most of such video-
based services, which is the expected case for more than 90% of future
Internet traffic. The pipeline forwarding parallel network is a “super-
highway” as it will carry a large part of the traffic with deterministic
performance as packets will flow faster and smoothly through it. Opti-
cal implementation of the Time-Driven Switching paradigm promises to
enable even more significant energy savings [74].
Figure 2.5. Parallel networks on the same fiber infrastructure.
32
2.4.4 Green Routing
In core networks, energy-aware routing is proposed as a novel routing
scheme, which uses energy consumption of network equipment as the
optimization objective. The authors in [61] propose an energy-aware
routing scheme which considers line card/chassis reconfiguration in IP
routers. Compared to the traditional shortest-path or non-energy-aware
routing scheme, energy-aware routing is expected to save a large amount
of energy. This is because line cards and chassis are major energy con-
sumers in core network and they are not configured and utilized energy-
efficiently in traditional routing schemes. In this energy-aware routing
scheme, energy consumption of IP routers in core networks is minimized.
In addition, future energy-efficient routing schemes may tend to be more
dynamic, which can reroute the traffic and save energy according to the
traffic variation during the day or the season. A study on how to adapt
OSPF (Open Shortest Path First) to include this kind of green routing
feature can be found in [75].
While energy efficiency may be part of the solution, recent research
[63] has also raised the concern that, given the rate of growth in demand
for ICT products and services, an increase in efficiency will not be suf-
ficient to counterbalance the growth in the ongoing deployment of new
equipment and services. As well, the tendency of users to increase con-
sumption of goods (in our case, energy) when the price of these goods
decreases (phenomena referred to as the Khazzoom-Brookes postulate
[76] or Jevons paradox) may mitigate any efficiency gains, i.e., it has
been demonstrated that, paradoxically, increased efficiency results in
increased consumption. So, depending solely on increased equipment
efficiency may not result in any significant reduction in GHG emissions
from computers and network equipment.
Under this perspective, since the target is essentially to reduce the
33
carbon footprint, we can devise approaches to decrease energy consump-
tion, targeting directly the reduction of GHG, which can help to solve the
Global Warming and related environmental problems. Therefore, renew-
able energy has gained more attention these days. An idea to reduce the
carbon footprint is to establish core servers, switches, and data centers
at locations where renewable energy can be found, and then to route the
traffic to the “Green areas” [64]. Since many network elements which
consume energy will be deployed at the locations of renewable energy,
zero carbon footprint can be realized. In this case, elements from other
part of the network may have to request the equipment in “Green areas”
to transfer their traffic demand by remote control, as shown in Fig. 2.6.
This approach sets up a connection between the energy-efficient network
and renewable energy utilization, which should gain more research in-
terest in the near future.
Figure 2.6. Green routing with availability of renewable energy.
Finally, traffic grooming is considered as a key functionality of WDM
networks, in which, multiple low-speed traffic requests are groomed onto
a high-capacity lightpath (wavelength) [19]. Energy-aware traffic groom-
ing approaches may also help to reduce the energy consumption of an
optical core network. Since network equipment consume a considerable
amount of energy even without any traffic flow [61] and the energy con-
34
sumption of most types of switching and transmission elements depends
on the traffic load to a certain extent, energy-aware traffic grooming can
be an approach to optimize the energy consumed by network elements.
In [65], total energy consumption of an optical WDM network is mod-
elled in terms of the energy consumed by individual lightpaths. Then,
an ILP (Integer Linear Program) formulation of the energy-aware groom-
ing problem is defined. Due to computational complexity, numerical so-
lution of the formulation is based on a small network, which indicates
that significant energy savings can be achieved with energy-efficient traf-
fic grooming. In [69] and [70], the authors propose both an MILP and
a heuristic approach to do routing and wavelength assignment to min-
imize the number of interfaces of lightpaths to minimize their energy
consumption. In [66] and [67], the authors consider energy consumed
by network operations while grooming traffic in optical backbone net-
works. Energy consumption of every operation in traffic grooming is
investigated, and an auxiliary-graph based model is proposed to iden-
tify the energy consumed by the operations. Results show that energy-
aware traffic grooming saves a significant amount of energy compared to
the traditional traffic grooming scheme. Authors in [68] also present a
traffic engineering scheme based on the idea that traffic grooming at the
lightpath layer can improve the energy efficiency of the network. They
studied how multilayer traffic engineering affects energy efficiency, and
their rationale is that the IP/MPLS (Internet Protocol/Multi Protocol La-
bel Switching) processing is more energy consuming than the lightpath
(optical) layer. In [71] and [72], the authors focus on energy-aware dy-
namic traffic grooming problem in optical networks. Based on the traffic
profile variation during different hours of the day, the authors minimize
energy consumption of the devices in the network.
Table 2.2 shows the comparison of greening efforts in core networks.
35
We compare the existing works in terms of the types of algorithms, energy
cost of the network, necessity of retrofit (whether network architecture
needs to changed), degree of energy savings, and extra signalling and
control.
2.5 Access and Metro NetworkIn this section, we review the research contributions on energy conser-
vation in access and metro networks. Most of the work in these areas
deal with access networks - some preliminary investigations on metro
network will be discussed at the end of this section.
A recent estimation [15] shows that access networks consume around
70% of overall Internet energy consumption. Hence, reduction of energy
consumption in access networks will lead to significant Internet energy
consumption reduction.
As bandwidth demands increase, access networks are becoming more
heterogeneous in nature as different access technologies are being com-
bined together. For example, current versions of xDSL use fiber as back-
haul, and CM access networks use HFC (Hybrid Fiber Coax) technology
as the network plant. Hence, developing energy-efficient fiber access
technologies will lead to future energy-efficient access networks. In this
section, we review the research efforts and recommendations aimed to
build energy-efficient wired (fiber and other) access networks.
The wireless networking community has been developing energy-effici-
ent wireless technologies for quite some time as extending the battery life
in a wireless device is a very important problem. These research efforts
can be summarized as a separate survey. In our work, we mainly focus
on optical networking technologies for energy-efficient access networks.
36
2.5.1 Energy Consumption Estimation
There are several publications which provide approximate estimations of
energy consumption in different types of access networks. The authors
of [3] present a basic energy-consumption model for generic access net-
works. They use the model to compare the energy consumption of point-
to-point optical links, PON, FTTN (Fiber To The Node), and WiMAX.
The efficiency of an access network can be defined as the energy con-
sumed per bit of data transferred [3]. In fiber-based access networks,
energy per bit drops as the average data rate increases. The per-user
energy consumption data shows that, for access rates below 300 Mbps,
PON is the most energy-efficient access network. Access networks with
FTTN and VDSL technologies (where per-user data rate is limited to 100
Mbps) consume two to three times more energy than PON due to the
presence of active remote nodes in the plant. WiMAX has the highest
energy consumption among all these access technologies at access rate
above 1 Mbps, and its date rate is limited to around 20 Mbps per user.
For data rates above 300 Mbps, the point-to-point fiber access network
becomes more energy efficient compared to PON as statistical multiplex-
ing gain in PON does not apply anymore. Hence, it is concluded that
PON and point-to-point optical networks are the most energy-efficient
access alternatives.
The authors of [77] extended the energy-consumption model of [3] and
studied the energy consumption of different FTTx network variants with
respect to the average access bit rate. Their results also conform with
the findings in [3] - up to a certain data rate, PON-based FTTx networks
are more energy efficient than point-to-point FTTx networks, and after
that rate, point-to-point FTTx networks are more energy efficient.
The authors of [78] measure the energy consumption of a content de-
livery network such as an IPTV network. They develop a simple energy-
37
consumption model for IPTV storage and distribution. This model can
provide guidelines for energy-optimized IPTV network design. It is sug-
gested that, for reducing energy consumption, frequently-downloaded
materials should be replicated at many data centers near the users and
less-frequent materials should be kept in a few data centers.
2.5.2 Energy-Aware Access Networks
In the previous subsection, we summarized results from publications
which estimated the energy consumption of access networks. Now, we
focus on different recommendations and research ideas on developing
energy-efficient access networks.
2.5.2.1 PON
There are two popular variants of PON – (a) EPON (Ethernet PON), which
uses Ethernet as the underlying transport mechanism, and (b) GPON
(Gigabit PON), an evolution of Broadband PON (BPON) standard. While
GPON standard is popular in Europe and North America, EPON domi-
nates the huge market in Asia. At the system level, PON technologies are
being improved for energy efficiency by (a) improved IC (Integrated Cir-
cuit) technologies such as smaller silicon size, (b) better devices such as
burst-mode laser drivers, (c) energy-efficient chips which shut down in-
active functions on the fly such as smart embedded processors, etc. [79].
Although neither PON standard incorporated any energy efficiency fea-
tures initially, after several proposals and deliberations, there are some
recommendations on building energy-efficient EPON and GPON. Below,
we give an overview of these recommendations. Although these rec-
ommendations are written separately, they can be incorporated in both
standards.
38
Tabl
e2.
3.C
ompa
riso
nof
gree
ning
effor
tsin
PON
.
App
roac
hSt
anda
rdiz
ed/
Prop
osed
Lega
cyC
ompl
iant
/
New
Arc
hite
ctur
e
Ext
raSi
gnal
ling
and
Con
trol
Impl
emen
tati
on
Com
plex
ity
Low
-pow
erst
ate
forO
NU
[80]
,[81
],[8
2],[
83]
Prop
osed
New
No
Mod
erat
e
Han
dsha
king
prot
ocol
for
coor
dina
ted
slee
ping
[81]
,[83
],[8
4]
Prop
osed
Com
plia
ntYe
sM
oder
ate
Shed
ding
pow
erin
UN
I
[79]
Stan
dard
ized
New
No
Mod
erat
e
Shed
ding
spee
dof
UN
I
[79]
Prop
osed
New
Yes
Hig
h
Shed
ding
pow
erin
AN
I
[79]
Prop
osed
New
No
Mod
erat
e
Shed
ding
spee
dof
AN
I
[79]
,[83
]
Prop
osed
New
Yes
Hig
h
39
• EPON: Current IEEE 802.3ah/802.3av EPON standards do not de-
fine any low-power state for the optical components such as OLT
or ONU [80]. However, during IEEE 802.3av task force meetings,
proposals have been circulated to include low-power states for ONU
so that it can go to sleep during network idle time [80]. It is esti-
mated that, during sleep state, power consumed by an ONU is at
least 10 times less than an active ONU [80]. Hence, there is a sig-
nificant scope of energy savings by putting idle ONUs to sleep. A
proper handshaking protocol is needed to arrange this coordinated
sleeping while not impacting service quality. In [84], the authors
propose such an adjustable-timer-based multi-point handshaking
protocol. Authors of [83] propose two energy-saving mechanisms
for 10G-EPON - one is sleep control function which switches modes
(active or sleep) of ONU depending on traffic variability, and the
other is an adaptive link-rate mechanism which switches the link
rate between OLT and ONU to conserve power.
• GPON: It is possible to shed power in the UNI (User Network In-
terface) (which connects ONU to user equipment) by turning it off
when not in use. This process is described in G.983.2 and G.984.4
recommendations and is supported by some existing products [79].
However, it is difficult to detect when the UNI is not active as con-
nected devices (such as computers) always communicate. It is also
possible to slow down UNIs that are not used fully, a process known
as UNI speed shedding [79]. Throttling back UNI speed in a seam-
less way can however be challenging.
We can also save energy by power shedding in the ANI (Access Net-
work Interface) which connects ONU to OLT. This technique basi-
cally turns off the whole ONU. It may have huge service quality im-
pact and may block incoming calls. Another technique can be ANI
40
speed shedding, i.e., slowing down the PON during low utilization.
This technique can be very complex to implement. Coordinated
scheduling of ONU shutdown based on time of the day can also
be explored for building an energy-efficient PON [79]. Implemen-
tation of sleep mode in GPON is described in ITU-T G.su45 GPON
power conservation standard [85]. Some GPON products have al-
ready included the power-saving mode which reduces up to 95%
of the ONU power consumption during power outages and standby
periods [86].
In [81], the authors present several power-saving modes for a TDM-
PON ONU and their advantages and disadvantages. They present a ONU
sleep-mode system architecture. A sleep-mode control protocol has also
been described in the paper. The authors of [82] demonstrate how sleep
mode can be realized in a TDM-PON ONU and energy can be conserved.
Once incorporated, the above techniques can save energy for both the
PON standards. Table 2.3 summarizes the comparison of the greening
efforts in PONs on the basis of standardization efforts, network archi-
tecture, degree of energy savings, requirement of extra signalling and
control, and implementation complexity.
2.5.2.2 xDSL
xDSL is the most dominant broadband access technology in the USA,
where 66% of the customers use DSL for accessing the Internet today [87].
One of the main communication challenges in xDSL is reducing electro-
magnetic interference known as crosstalk which occurs due to signal
interference of different lines in the same cable bundle. Crosstalk can
hugely deplete the DSL line’s available bandwidth, and by decreasing
crosstalk, it is possible to increase the operating efficiency and energy
efficiency of DSL lines.
There are two different ways for reducing the crosstalk in DSL lines:
41
(1) Smart DSL and (2) DSM (Dynamic Spectrum Management). Smart
DSL is a proprietary technology developed by Alcatel-Lucent which in-
troduces low-level noise in DSL lines to mask the crosstalk [88]. One
can also combine Layer-2 Power Mode with smart DSL to improve en-
ergy efficiency of ADSL2+ deployments. This combination cancels out
power fluctuations, decreases crosstalk, and creates a more stable net-
work [88].
The other alternative – DSM – curbs crosstalk rather than masking
it out. DSM coordinates the spectrum and/or signals from all users
to reduce crosstalk [89]. Regular DSM design can be extended to add
constraints on transmit power so that overall power consumption by DSL
lines gets minimized [89]. Low transmit power will eventually reduce the
power consumed by DSL modems. Low transmit power will also lead to
less crosstalk between DSL lines. All of these features can be combined
to make DSL “green” and energy-efficient [90].
It is estimated that there are opportunities for up to 50% energy sav-
ings while achieving 85% full-power data rate performance in real DSL
network scenarios [89]. There are several solutions for reducing trans-
mission power in DSL systems such as adaptive startup and L2 mode
[91]. Implementations of constrained maximum transmission power and
modes exploiting traffic-dependent transmission power are also being
considered [91].
2.5.2.3 WOBAN
WOBAN is a proposal for an optimal combination of an optical backhaul
(e.g., PON) and a wireless front-end (e.g., WiFi and/or WiMAX) [4]. In
WOBAN, a PON segment (headed by OLT) starts from the telecom CO
and serves several ONUs. One ONU can serve several wireless gateways
which, in turn, gather traffic from the wireless mesh front-end. There
is a capacity mismatch between the wireless front-end and the optical
42
backhaul. The extra capacity in the optical backhaul not only serves
regular PON traffic but also provides enhanced reliability during a net-
work failure so that traffic can be rerouted through alternate paths in
the wireless front-end. This flexibility provided by the wireless front-end
can be exploited during low-load hours to enable energy savings in the
optical part of WOBAN (please see Chapter 4).
Traffic load on an access network fluctuates at different hours of the
day. During low-load hours, the under-utilized part of WOBAN can be
put to sleep while rerouting the affected traffic through other parts of
the network. For the wireless front-end of WOBAN, coordinated sleep-
ing techniques from mobile ad-hoc networks research can be adopted
to reduce wireless router energy consumption. For the optical part, the
OLT can manage a centralized sleeping mechanism to put low-load ONUs
to sleep (details in Chapter 4). To reroute the affected traffic while not
impacting the service quality, an energy-aware routing algorithm is de-
vised in Chapter 4. The objective of the routing algorithm is to “use the
already-used paths” while keeping the average path length comparable
with shortest-path routing.
2.5.2.4 Long-Reach PON
LR-PON (Long-Reach PON) is proposed as a cost-effective solution for
future broadband optical access networks. LR-PON extends the coverage
span of PONs (from traditional 20 km range) to 100 km and beyond by
exploiting Optical Amplifier and WDM technologies [26]. In this way,
LR-PON consolidates several remote central offices into a central one,
thereby reducing the energy usage of future access networks. In LR-
PON, each PON segment has the traditional tree topology, and the OLT is
connected to those PON segments by a fiber ring and remote nodes (RN).
The authors of [92] present a dynamic wavelength allocation scheme for
LR-PON. This scheme assumes wavelength sharing among several RNs
43
and reduces energy consumption of LR-PON by minimizing wavelength
requirements and putting idle transmitters to sleep.
2.5.2.5 Energy Conservation in Metro Networks
There is limited research on energy conservation in metro networks. The
authors in [93] deal with energy-efficient design of network architectures
for metro networks. They consider three architectures for a unidirec-
tional WDM ring network, i.e., FG (First-Generation) optical network,
SH (Single-Hop) network, and MH (Multi-Hop) network. In a FG optical
network, every node must electronically process all the incoming and
outgoing traffic, including the in-transit traffic. In a SH optical network,
every node electronically processes only the traffic that goes into or out
of the network at that node. A MH network lies somewhere between the
FG and SH networks.
The MH architecture makes use of both lightpaths and electronic
traffic multiplexing, performed at a few selected intermediate nodes. A
power-saving network design is proposed, aiming at minimizing the en-
ergy required by both optical and electronic components. The energy
consumption for the three architectures is optimized using ILP formula-
tions. The authors show that, when the unidirectional WDM ring net-
work has uniform traffic, the power consumption of the MH network is
lower than that of the FG network, not only when traffic load of optical
components is low, but also when connection rate is close to the wave-
length capacity. The authors also show that, when the connection rate
is low, the MH network outperforms the all-optical SH network, because
the MH network has more flexibility to perform traffic multiplexing in an
energy-efficient way.
44
2.6 Data Centers and Applications2.6.1 Data Centers
Data centers are vital to support many of today’s data-intensive telecom
applications. The huge amount of data to be managed by these applica-
tions has been posing scalability issues for the data center infrastruc-
tures, and optical technologies represent a key enabler for data centers
to support all of these traffic.
Specifically, optical networks play a relevant role in both data center
inter- and intra-connections. At the inter-connection level, moving and
delivering the ever-increasing amount of traffic to be supported by data
centers can be effectively done using reconfigurable optical networks:
note that the flexibility of the inter-connection pattern of the core trans-
port network, which is promised to be provided by emerging automatic
control plane suites such as GMPLS (Generalized Multi-Protocol Label
Switching) and ASON (Automatically Switched Optical Network), will be
an important means to transfer data load among various sites, as en-
visioned in most of the works which are reviewed in this section [94],
[95].
At the intra-connection level (connecting boards, chips, and memo-
ries of the data servers inside the data center), optical technology can
also play a fundamental role for data center scalability: optics could
solve many physical problems of intra-connections, including precise
clock distribution, system synchronization (allowing larger synchronous
zones, both on-chip and between chips), bandwidth and density of long
interconnections, and reduction of power dissipation. Optics may relieve
a broad range of design problems, such as crosstalk, voltage isolation,
wave reflection, impedance matching, and pin inductance. It may al-
low continued scaling of existing architectures and enable novel highly-
connected or high-bandwidth architectures [96].
45
Since servers and associated equipment consume a considerable part
of energy used in telecom networks, several recent studies have focused
on the estimation of the energy consumption in data centers. As an ex-
ample, the total power used by servers in data centers represented about
0.6% of the total U.S. electricity consumption in 2005. When cooling and
auxiliary infrastructure are included, this number grows to 1.2%, which
is an amount comparable to that for televisions [97]. Therefore, energy-
conservation technologies for data centers are being developed.
The author in [98] has proposed an approach for power control of
high-speed network intra-connection (inside data centers), which focus
on reducing the energy consumption of communication links. The au-
thor claims that communication links can support three types of power
control: (i) usage of one or more low-power states, (ii) link width con-
trol, where only a portion of the link is put into a low-power mode, and
(iii) multiple operational speeds [98]. The author focuses on method (ii).
The width control algorithm decides how to transit between certain fea-
sible widths in a multilane link, which involves energy-efficient design
of networking fabrics, as well as interconnects that proliferate inside a
server, e.g., CPU core interconnects, processor-memory interconnects,
PCI-E (Peripheral Component Interconnect Express) links connecting
NICs (Network Interface Controllers), graphics card, SATA (Serial Ad-
vanced Technology Attachment) adapters, etc. Results show that, when
link width grows but traffic demands stay the same, power consump-
tion can be brought down after power control. This is because links with
higher width have higher probability of holding spare resources than the
ones with lower width.
Another aspect of power-conservation technologies in data centers is
about load distribution across data centers in different locations. A
framework for optimization-based request distribution is proposed in
46
[94]. Leveraging the combination of different time zones (where differ-
ent data centers may be located), variable electricity prices, and some
data centers being powered by green energy, an optimal load-distribution
scheme across data centers is proposed. Mathematical optimization for-
mulations and heuristics are proposed to minimize the cost and energy
consumption of the collection of data centers. Since traffic demands
vary at different locations during time of the day, after a specific request
distribution, energy and cost can be minimized by the energy-efficient
framework. This approach also provides a novel way to better utilize re-
newable energy.
Along the lines of the previous concept, another approach for en-
ergy conservation based on traffic load redistribution consists in locating
servers at sites where renewable energy is available and then connecting
these servers with the rest of the network by using optical transport sys-
tems. As an example, considering location availability of renewable en-
ergy, some institutions are about to launch a $100M “green” data center
in the city of Holyoke, where there is a ready source of cheap, relatively-
clean hydroelectric power [99]. This project promises to be very helpful to
reduce the carbon footprint of data centers in the eastern United States.
Google’s “project 02” and Microsoft are also using hydroelectric facilities
to build data centers to utilize renewable energy [95]. IBM, Syracuse Uni-
versity, and New York State have entered into an agreement to build and
operate a new data center on the Syracuse University’s campus. They
will incorporate advanced infrastructure and smarter computing tech-
nologies to make it one of the most energy-efficient data centers in the
world. The data center is expected to use 50 percent less energy than a
typical state-of-the-art data center. The key element is an on-site electri-
cal co-generation system that will use natural-gas-fuelled micro-turbine
engines to generate all the electricity for the center and provide cooling
47
for the computer servers [100]. On this topic, still a lot of research is
needed on devising new Internet architectures with servers, computers,
and storage collocated at remote renewable energy sites such as hydro
dams, windmill farms, etc. Also, new routing and protection strategies
for optical networks are sought for rapid and massive network-wide re-
configuration of the network interconnection between data centers ac-
cording to current availability of renewable (e.g., sun or wind) energy to
power routers and servers [64].
In the management of data center networks, a single administrative
control domain is proposed for energy conservation of data centers [101].
The authors envision a centralized network power controller program
running on a server within the data center. The energy-efficient algo-
rithms can be link-state adaptation, network traffic consolidation, and
server load consolidation. In these schemes, the placement algorithms
take network traffic specifications of the job, the current network uti-
lization, and the connectivity into consideration before assigning var-
ious servers for a job. Then, the controller communicates with all the
switches and performs actions such as turning off unused switches, dis-
abling unused ports, and adapting link capacity to save energy.
The authors in [101] also propose a power benchmarking framework
for network devices in data centers. They build and describe a bench-
marking suite that will allow users to measure and compare the power
consumed for a large set of common configurations in any switch or
router of their choice. They also propose a network energy proportional-
ity index to compare power consumption behaviors of multiple devices.
In their scheme, the network device to be benchmarked is connected to
the power outlet via a power meter. Then, the device configurator modi-
fies the various configuration states of the device according to the bench-
marking requirements. The traffic generator loads the device with vary-
48
ing traffic patterns. The benchmark orchestrator coordinates the various
components in order to synchronize the configuration, the workload, and
the measurements from the power meter. The collected information is
then processed by an analyzer to generate various energy proportionality
indices and other power-related metrics [102].
2.6.2 Applications
While storage, memory, processor, and communication bandwidth tend
to become relatively abundant and inexpensive as time progresses, elec-
tricity usage will become a growing expense in the operation of telecom
networks [103]. In the application layer of computers and, more gen-
erally, telecom networks, turning idle devices to sleeping mode appears
to be the most plausible way in which energy conservation can be well
achieved. However, in order to implement algorithms for sleeping, sev-
eral aspects have to be considered, e.g., (i) software should be designed to
enable hardware of network equipment to sleep, (ii) Internet routing pro-
tocols, such as TCP/IP, need to be modified to adapt to energy-efficient
design, and (iii) hardware of network equipment needs to be reconfigured
to accept control signals from the software [104]. Several approaches
have been identified that satisfy the above requirements, and they target
energy conservation at the application layer. Broadly, we can identify
three main proposals: “Proxying”, Green TCP/IP protocol, and Green
Grid Computing.
Below, the first two areas of research are quickly outlined for the sake
of completeness, since they are not specifically related to optical network
technologies. A longer discussion will be provided on green grid com-
puting because of its close relation to optical networks. In fact, the com-
putational resource sharing and virtualization enabled by optical grid
networks (also referred to as lambda grids) is raising a lot of interest as
practical means to reduce energy consumption. In [105], the authors an-
49
alyze the energy-saving opportunities of the thin client paradigm where a
thin client terminal (with less functionalities) consumes less power, and
more efficient use of resources in the server is possible due to virtual-
ization.
2.6.2.1 Proxying
A first possible approach for reducing power consumption at application
layer consists of using network connectivity “Proxying”. Since much of
the network connectivity should be maintained at all times to allow re-
mote access and/or operations of network-centric applications, the PCs
and servers involved have to be kept always on (day and night). In this
case, a large amount of energy will be consumed. However, these PCs
and servers are probably idle for significant durations of time. The au-
thors in [17] propose a “Proxying” scheme that enables idle PCs to use
sleeping mode. The structure of this “Proxy” scheme is shown in Fig.
2.7.
Figure 2.7. Network connectivity “Proxying”.
When a PC becomes idle, it transfers its network presence to the proxy
before going to sleep, and then the proxy responds to route network traf-
fic for the sleeping PC. When the PC is needed, the proxy wakes it up. In
this case, the energy consumption of the system can be reduced because
the proxy consumes much less energy than the monitor, hard disk, or
50
CPU of a PC does. At the same time, TCP connections can be kept alive
during the PC’s sleep period by using a SOCKS-based (Protocol for ses-
sions traversal across firewall securely) approach called green SOCKS
(gSOCKS) as part of the Network Connectivity “Proxying” [27].
2.6.2.2 Green TCP/IP Protocol Design
At the application layer, protocols for IP routing determine the opera-
tional performance of the network to some extent, such as transmission
delay or energy consumption. Many PCs are kept on in corporate offices
at night, even when no applications or network activities are running on
them, while in residential areas, many people keep their PCs on when
they leave their house for work or holiday. In this way, a large amount of
energy is wasted. In [106], a green TCP/IP protocol is proposed, which
enables existing TCP/IP connections to be “put to sleep” to save energy.
The green TCP/IP protocol also helps servers to block network connec-
tions between servers and clients when client PCs are sleeping. Network
connections can automatically resume when the client PCs wake up.
Figure 2.8. Grid Computing job scheduling mechanism.
2.6.2.3 Green Grid Computing
Grid computing combines computing resources from multiple adminis-
trative domains for a common goal [107]. Distributed grid computing,
in general, is a special type of parallel computing that relies on com-
51
puters connected to a network. Grid computing was originally started
for Internet-related services such as search engines. Today, many other
services, applications, and tasks that used to reside on an end user’s
terminal or computer get transferred to the grid. Software such as Sun
Grid Engine, GridWay, etc. are developed to meet the requirements of
next-generation grid computing. The underlying network architecture
building the foundation for grid computing consists of interconnected
server farms within data centers and a high-speed transport network
providing connectivity to remote and backup sites. These high-speed
connections form the backbone of the grid network and are required
to run at highest bandwidth with lowest transmission latency - in par-
ticular, high-speed grid optical networks, such as National LambdaRail
[108], promise to revolutionize the way that we approach grid comput-
ing, providing a scalable, reconfigurable, and cost-effective platform to
support grid services [109].
In recent years, grid computing is also dealing with large experimen-
tal bulk data obtained from large-scale scientific instruments (e.g., radio
telescopes used in the VLBI (Very Long Baseline Interferometry) exper-
iments), high-end physics experiments at CERN (European Organiza-
tion for Nuclear Research), or large-scale data processing results. In
order to meet these huge computational and storage demands, com-
putational cluster centers (e.g., supercomputers) are interconnected via
networks to achieve a huge common resource pool to process the tasks
[28]. Grid-based applications are also the hallmark of the twenty-first
century global e-Science, which is defined as global, large-scale scien-
tific collaborations enabled through distributed computational and com-
munication infrastructure. In [110], the authors reviewed related open
research issues on optical network control plane for the grid community
to meet the requirements of high-bandwidth connectivity for supporting
52
high-end supercomputers and highly dynamic operation. GMPLS-based
Traffic Engineering is also proposed in [111] to analyze the performance
of infrastructure service provisioning. The results show that the majority
of performance improvements (such as efficiency of resource utilization)
can be obtained with a controlled usage of multi-layer resource visibility
and with a more flexible interconnection architecture between domains.
Load balancing is also a crucial issue for the efficient operation of grid
computing environments in distributing the sequential tasks. The au-
thors in [112] propose a novel combination of static and dynamic load-
balancing strategies which helps to reduce the system response time and
to perform rapid task assignments.
As grid computing is being widely investigated in recent research,
power-aware grid computing schemes have also been proposed. Recent
studies of the usage of grid resources shows that the usage of a grid site
may significantly vary (between less than 20% to over 90%) during the
time of day [113]. Therefore, there is an opportunity for using energy-
saving mechanisms to automatically switch on and off servers to match
the available server capacity to actual computational demands. In grid
systems, users do not really care about where exactly their jobs ulti-
mately get executed; the job can be offloaded to a remote site with an
available processor, rather than turning on a new server, which can re-
duce energy consumption of the whole grid system. To reduce energy
consumption, a grid system needs a power-aware job scheduling mecha-
nism, and a power-saving strategy to decide when to turn servers on/off.
As Fig. 2.8 shows, the job scheduling mechanism first considers the
servers which were already powered on (server 2 or 3). Only if none is
available, the mechanism then turns on one among those servers which
were powered off using a shortest-path strategy (server 4 or 5). To decide
when to turn servers off, a straightforward approach is proposed: every
53
server will be turned off for a fixed time D after a job is finished, if during
that time it is not running any other job or being used as the intermediate
server for other working servers [28]. In this way, grid computing will not
be interrupted when idle servers are turned off during the computation,
and a large amount of energy will be saved during the time idle servers
are turned off.
2.7 ConclusionEnergy efficiency in telecom networks is a recent research topic, but it
is gaining rapid recognition in the research community, motivated by
the concern for the ever-increasing energy consumption of ICT. This sur-
vey reviewed energy-conservation protocols and energy-efficient archi-
tectures over the different domains of telecom networks, namely core,
metro, and access networks, with a specific emphasis on telecom net-
works employing optical technologies. Important applications running
over optical networks such as grid computing and data centers net-
works were also considered. Besides, standardization efforts toward en-
ergy efficiency by various telecommunication organizations were sum-
marized, which may provide practical references to researchers. We
provided useful references for researchers interested in energy-efficient
telecom networks, which can be helpful to develop future directions on
“green-networking” research.
54
Chapter 3
WOBAN Prototype Developmentand Research Challenges
3.1 IntroductionDuring the past decade, the backbone network has experienced enor-
mous growth in capacity and reliability, mainly due to major development
efforts in the area of optical networking. During the same time, band-
width demands of technology-savvy end users for broadband services
such as “quad-play” (voice, video, Internet, and wireless) and media-rich
applications have also increased at an unprecedented rate. However,
the access network (commonly referred to as the “last-mile” network)
still remains a bottleneck for providing bandwidth-intensive services to
customers. Legacy access technologies (such as Digital Subscriber Line
(DSL) and Cable Modem (CM)) will not be able to carry the high volume
of traffic generated by emerging applications such as video-on-demand
(VoD), interactive gaming, or duplex video-conferencing. Thus, future
access technologies should provide high capacity and operational effi-
ciencies along with mobility support and untethered access to users in
a cost-effective manner.
Optical-fiber-based technologies (e.g., fiber-to-the-home (FTTH), fiber-
to-the-building (FTTB), fiber-to-the-curb (FTTC)) are well suited to sup-
55
port integrated high-bandwidth digital services, and can alleviate band-
width bottlenecks. The next generation of access networks is therefore
promising to deploy optical fiber all the way to the customer premises.
However, laying fiber infrastructure to all end-users incurs significant
cost. Furthermore, users also desire untethered access, especially if
they are mobile. Wireless technologies can support mobility and unteth-
ered access. Unfortunately, wireless access is constrained due to limited
bandwidth. Therefore, combining the complementary features of these
two technologies (optical and wireless) can potentially provide ubiqui-
tous (“anytime-anywhere”) broadband access to satisfy future customer
demands. Therefore, a novel cross-domain network paradigm – Wireless-
Optical Broadband Access Network (WOBAN) – which is an optimal com-
bination of high-capacity optical backhaul and untethered wireless ac-
cess, is proposed in the literature [114].
WOBAN shows excellent promise for future access networks. This
cross-domain network architecture consists of an optical backhaul (e.g.,
a Passive Optical Network (PON)) and wireless access in the front-end
(e.g., WiFi and/or WiMAX). In WOBAN, a PON segment starts from the
telecom Central Office (CO) with an Optical Line Terminal (OLT) at its
head end. Each OLT can drive several Optical Network Units (ONU),
and each ONU can support several wireless routers of the wireless front-
end in WOBAN. The wireless routers directly connected to the ONUs are
called wireless gateways. The wireless front-end also consists of other
wireless routers to provide end-user connectivity. Therefore, the front-
end of a WOBAN is effectively a multi-hop Wireless Mesh Network (WMN)
which is connected to the high-capacity PON segment in the back-end,
creating a cross-domain integrated network architecture.
There is another related architecture, known as Radio-Over-Fiber (ROF),
where radio signals can be effectively carried over an existing optical fiber
56
infrastructure using “Hybrid Fiber Radio” (HFR) technology [115]. ROF
deals with the communication challenges of sending radio signals over
fiber whereas WOBAN focuses on the networking aspects of the wireless-
optical converged architecture.
In this chapter, we present the experiences gathered during a WOBAN
prototype development, and discuss future research issues to improve
the performance and design of this hybrid network. We provide detailed
prototype development procedures and introduce some of the challenges
involved in the development. The WOBAN prototype serves as the exper-
imental setup for various access network protocols and data dissemina-
tion techniques; and it features programmability, resource sharing, and
slice-based experimentation. We believe that this prototyping effort will
lead us to identify and address several practical concerns that WOBAN
may encounter in future. This prototype will also enable researchers to
study and evaluate energy-conservation mechanisms for WOBAN in real
networking environment.
The remainder of this chapter is organized as follows. We first present
related prototyping efforts on hybrid cross-domain networks in the lit-
erature. We then present the WOBAN prototype architecture, its dis-
tinguishing features, and its development procedure. Experimental re-
sults are demonstrated and discussed in the following section. Then, we
elaborate on future research challenges of WOBAN. Finally, concluding
remarks are provided.
3.2 Related Development EffortsThis section briefly reviews other testbeds/prototypes developed for hy-
brid wireless-optical networks research.
Hu et al. [116] have developed a testbed for an Optical-Wireless Inte-
gration (OWI) infrastructure. They implemented SONET/WDM, popular
57
in core optical networks, for the optical part and WiMAX (IEEE 802.16)
for broadband wireless access. The edge node between two networks in-
terfaces the WiMAX base station and SONET with a direct conversion
between the protocol stacks of the optical and wireless segments.
Grid Reconfigurable Optical and Wireless Network (GROW-Net) [117]
is another hybrid wireless-optical network which consists of an “Infras-
tructure” based WMN in the front-end and a reconfigurable, high-capacity,
point-to-multipoint PON optical backhaul. To demonstrate the perfor-
mance of the proposed optical backbone reconfiguration scheme in GROW-
Net, the authors of [117] developed only an optical experimental testbed
based on commercially-available devices. This testbed is dedicated to
optical backhaul reconfiguration experiments.
Jia et al. [118] have developed a testbed for Radio-Over-Fiber (ROF)
experiments. The testbed has two segments – (a) Central Station (CS) and
(b) Base Station (BS) – and it consists of optical transmission equipment.
The main purpose of this testbed is to illustrate how wireless signals can
be carried over fiber. This testbed demonstrates the feasibility of a full-
duplex ROF system based on optical carrier suppression and reuse for
future optical/wireless networks.
3.3 Implementing WOBAN PrototypeIn this section, we discuss the logistics (resources needed for prototype
development), WOBAN architecture, features, and detailed development
procedure.
3.3.1 Resources Needed
Table 3.1 summarizes various device specifications used in our proto-
type. All these devices are commercially available off-the-shelf devices
and can be used effectively to build a fully-functional and reasonable-
sized prototype.
58
Table 3.1. WOBAN prototype components and their specifications.
Components Interface/Port
OLT • Client Side: One EPON port
• Network Side: One 100/1000 Base-T Ethernet
port (for RoI (Rest-of-the-Internet))
ONU • Client Side: Two 10/100 Base-T Ethernet ports
(to drive 802.11g routers)
• Network Side: One EPON port (to connect OLT)
Optical Splitter 1:8 power splitter
802.11g Router • Client Side: One radio port
• Network Side: 10/100 Base-T Ethernet port
Clients Laptops, PDAs, etc.
We use open source firmwire OpenWRT1 to develop the reconfigurable
wireless routers and gateways.
3.3.2 Architecture
Figure 3.1 shows the architecture of WOBAN prototype developed in the
Networks Research Laboratory at UC Davis.
The wireless routers form the WOBAN front-end and connect to the
end users (who can be scattered over the geographic area served by the
WOBAN and who are not shown in Fig. 3.1). These wireless routers
(IEEE 802.11g) support data rates up to 54 Mbps. Several designated
routers are configured to have Gateway capabilities (by loading appro-
priate open source firmware) and each such Gateway is connected to an
ONU via a 10/100 Base-T Ethernet port. The wireless routers are placed
with an effective distance of 50-60 meter between pairs.
1“OpenWrt”, http://www.openwrt.org/.
59
v v
v
OLT
1
ON
U 1
OLT
2
ON
U 2
ON
U 3
Fib
er
Fib
er
1:8
Opt
ical
S
plitt
er
100M
/1G
B
ase
T
100M
/1G
B
ase
T
RJ4
5 10/1
00
Bas
e T
10/1
00
Bas
e T
10/1
00
Bas
e T
RJ4
5
RJ4
5
Wire
less
Lin
k 54
Mbp
s
Gat
eway
2
Rou
ter
Rou
ter
Wire
less
Lin
k 54
Mbp
s
1:8
Opt
ical
S
plitt
er
Linu
x B
ox
Linu
x B
ox
Linu
x B
ox
Eac
h O
LT d
rives
diff
eren
t pa
rts
of W
OB
AN
E
ach
ON
U a
nd it
s at
tach
ed L
inux
Box
em
ulat
e th
e op
erat
ion
of a
n O
NU
W
irele
ss R
oute
rs h
ave
dist
ribut
ed c
ontr
ol: l
ink
form
atio
n an
d ro
utin
g
Wire
less
Rou
ters
and
G
atew
ays
are
conf
igur
ed d
iffer
ently
fo
r th
eir
resp
ectiv
e fu
nctio
nalit
ies
Con
trol
, Dat
a, a
nd M
anag
emen
t Pla
nes
at W
OB
AN
N
OC
Res
t of t
he In
tern
et
(UC
Dav
is N
etw
ork)
C o n
t r o l
P l a
n e
D a t
a P
l a n e
M a
n a g
e m e
n t P
l a n
e
Con
trol
ler
Por
tal
Man
agem
ent
WO
BA
N N
OC
Gat
eway
3
Gat
eway
1
WO
BA
N C
O
Figu
re3.
1.W
OB
AN
prot
otyp
ear
chit
ectu
re.
60
Two OLTs (Optical Line Terminal) emulate the functionality of the tele-
com Central Office (CO) of the general WOBAN architecture. Each OLT
can drive several ONUs using an optical splitter. The OLTs and ONUs
are connected through Ethernet PON (EPON) ports. The OLTs are con-
nected to the Rest of the Internet (ROI) using the campus-wide backbone
network at UC Davis.
The prototype architecture is divided into three planes: (a) Control
Plane, (b) Data Plane, and (c) Management Plane. The Control Plane
is used to define different control features of the nodes in the WOBAN
prototype. The Data Plane configures routing and different data-transfer
scenarios, and collects measurement data for different experiments. The
Management Plane is used for remote access and programmability of the
prototype nodes. The WOBAN Network Operations Center (NOC) (see Fig.
3.1) is responsible for the management of all these planes.
3.3.3 Distinguishing Features
The WOBAN prototype has several distinguishing features which are dif-
ferent from other related prototypes ([116], [117], [118]) reported in the
literature, as follows.
• To the best of our knowledge, this is the most integrated wireless-
optical hybrid network testbed. Other testbeds have only a small
number of nodes and have been used as proof of concepts. On
the other hand, WOBAN prototype features programmability, self
organization, and slice-based experimentation.
• The WOBAN prototype is large enough to demonstrate its useful
properties, e.g., two OLTs can demonstrate fault-tolerance prop-
erties of WOBAN so that, if one OLT breaks, the other parts of
the WOBAN can “self organize” themselves to still carry the af-
fected traffic through the other operational parts of the WOBAN.
61
The self-organization property of WOBAN also holds for (1) other
failure types, e.g., ONU failure, fiber cut, wireless router failure,
etc. and (2) optimal routing.
• The deployment and management cost of WOBAN prototype is low
as it is built from highly-customized off-the-shelf components, open
sources, and indigenous software.
• The front-end can be set up as a plug-and-play wireless mesh.
• The prototype nodes feature programmability. The open source
firmware provides the programmability in the wireless routers. The
programmability of OLT can be performed by using the craft port
in the OLT box and the ONU programmability can be emulated by
gluing a separate “Linux box” with each ONU.
• The prototype is reconfigurable and provides self-organizing and
self-healing properties. The reconfigurability is performed by Layer-
2 (L2) connectivity and intelligent routing.
• Power consumption of the wireless nodes is very low (1-2.5 Watts/
router). As the wireless mesh constitutes a large part of the proto-
type, the overall power consumption is also low.
3.3.4 Development Procedure
Here, we present deployment issues related to different planes in the
WOBAN prototype and show how they are addressed during the deploy-
ment phase.
3.3.4.1 Control-Plane Issues
• Topology Creation/Connectivity: The optical segment of the WO-
BAN prototype has a static topology initially as connections be-
tween nodes are wired. The wireless segment uses proactive rout-
ing (namely Optimal Link State Routing (OLSR) in our prototype) to
62
create a “self organizing” topology where, in case of a router failure,
nodes can redirect traffic to the nearby active routers. If a failure
occurs in the optical segment, dynamic protection scheme can be
applied for “self-healing”.
• Dynamic Bandwidth Allocation (DBA): The optical part of the
WOBAN prototype uses Ethernet PON (EPON) as the basic technol-
ogy. In EPON, the Ethernet functionality is emulated by a Layer-2
signalling mechanism, called Multi-Point Control Protocol (MPCP)
[2] that would allow the OLT to assign the bandwidth dynamically
among ONUs. We can use hierarchical MPCP-based protocol in two
levels (OLT-to-ONUs and ONU-to-Gateways) coupled with Layer-2
signaling (Gateways-to-Routers) for DBA, and thereby achieving str-
onger wireless-optical integration. Overview of this kind of protocol
is given in a later section.
• Programmability: An important aspect of the WOBAN prototype
nodes is their programmability. Experimental testbed researchers
should be able to create, modify, and test their protocols on the pro-
totype. In our WOBAN prototype, we create a simple remote-access-
based programmability platform for the wireless nodes (gateways/
routers). This platform provides programmability at each layer of
the IEEE 802.11 protocol stack. The OLT DBA mechanism (Layer-2
signalling) can also be programmed using the craft port installed
in the OLT box. For ONU, we can emulate the programmability by
gluing a “Linux box” with each of them.
3.3.4.2 Data-Plane Issues
• Routing: Proactive routing such as Optimal Link State Routing
(OLSR) is used in the wireless mesh and Layer-2 static routing is
used in the optical part of the WOBAN prototype. Dynamic rout-
63
ing protocols such as OLSR waste significant amount of wireless
bandwidth for periodic link-state updates. From our prototype ex-
perience, we find that static routing can perform better compared to
a dynamic approach in a WOBAN-type network architecture. One
such proposal is discussed below.
• Configurations: Prototype nodes can be configured for different
experiments. These data-transfer configurations facilitate us to ob-
tain experimental data for various applications on the WOBAN pro-
totype.
• Measurement: Network protocol analyzers (e.g., tcpdump, Wire-
shark2, etc.) are used to collect and analyze network statistics from
various experiments.
3.3.4.3 Management-Plane Issues
• Remote Access: In the WOBAN prototype, we use remote access
interfaces to download our own code inside the nodes and run the
experiments. Wireless nodes are connected with the Network Oper-
ations Center (NOC) through wireless interfaces, and optical nodes
are connected through craft ports.
• Network Slicing: To share the WOBAN testbed resources among
several experiments, currently physical slicing is used. In physi-
cal slicing, resources are physically divided among different experi-
ments. We can also implement the virtual slicing feature where the
physical resources of WOBAN nodes can be shared among exper-
iments. Time-Division Multiplexing (TDM) based virtual slicing is
very challenging to implement [119]. Further research is required
to deploy such features in the prototype.
2“Wireshark”, http://www.wireshark.org/.
64
OLT
1
ON
U 1
OLT
2
ON
U 2
ON
U 3
Gat
eway
1
Res
t of t
he In
tern
et
(UC
Dav
is N
etw
ork)
Gat
eway
2
Gat
eway
3
Wire
less
Li
nk W
irele
ss
Link
Opt
ical
B
ackh
aul
Wire
less
F
ront
-End
Dat
a ov
er 4
w
irele
ss h
ops
VoI
P o
ver
2 w
irele
ss h
ops
Vid
eo o
ver
3 w
irele
ss h
ops
V i d e o S e r v e r c u d a . c s . u c d a v i s . e d u
D a t a S e r v e r s e r r a n o . c s . u c d a v i s . e d u
VoI
P C
lient
VoI
P C
lient
S
plitt
er
Spl
itter
Vid
eo
Clie
nt
Dat
a C
lient
Figu
re3.
2.W
OB
AN
prot
otyp
eex
peri
men
tals
etu
p.
65
3.4 Experimental IllustrationsHere, we present experimental results collected from the WOBAN proto-
type for various applications (Data, Voice-over-IP (VoIP), and Video-on-
Demand (VoD)).
3.4.1 Experimental Setup
Figure 3.2 shows the setup for different experiments on WOBAN proto-
type. The wireless front-end of WOBAN should have distributed control
to exhibit self-healing and self-organization properties. Therefore, we use
IEEE 802.11 basic Distributed Coordination Function (DCF) for medium
sharing. IEEE 802.11 Point Coordination Function (PCF) is only suit-
able in wireless “infrastructure” mode, hence is not feasible for WOBAN
wireless mesh front-end. IEEE 802.11e-based enhanced coordination
functions for better QoS performance have not been considered in our
experiments as this standard is relatively new and is still in the develop-
ment phase. Transmission rate of wireless routers is set to 54 Mbps.
In all the experiments, background traffic load is generated using
software-based traffic generators. We run our experiments under no
background load to heavy background load to determine the effects of
background traffic on different applications. In all the experimental se-
tups, one end (server/client) of a connection is located in the RoI, and
the other end (wireless client) is connected to the wireless mesh through
multiple hops. Background traffic also flows between these two ends so
that all the links of a experimental connection experience some external
traffic load.
The quality of the wireless channels varies randomly across the ex-
periments due to different interference factors in our environment. This
inherent randomness of wireless channels may have impacts on accumu-
lated results. The impact of wireless channel quality on the performance
66
is not studied in these experiments. We mainly focus on various appli-
cations’ performance under random wireless environments. Our results
indicate that, as the number of wireless hops increases, various per-
formance quality measures decrease, due to bottleneck in the wireless
mesh. Therefore, our accumulated results present the performance of
different applications by varying the number of wireless hops.
3.4.2 Results3.4.2.1 Data
We start with data-transfer applications such as secure file transfer (viz.,
sftp or winscp). In our experiments, transferred file size is 76 MBytes.
Figure 3.3 shows the data-transfer application’s end-to-end through-
put. As expected, with increasing number of wireless hops, end-to-
end throughput decreases significantly. Furthermore, presence of back-
ground traffic decreases the throughput.
1 2 3 40
0.5
1
1.5
2
2.5
3
Wireless Hop Count
Thr
ough
put (
Mbp
s)
Throughput vs. Wireless Hop Count
Without Background Traffic Background Traffic 1.5 MbpsBackground Traffic 3.0 Mbps
Figure 3.3. Data-transfer throughput.
67
1 2 3 40
5
10
15
20
25
30
Wireless Hop Count
Pac
ket L
oss
Rat
e (%
)
Packet Loss Rate vs. Wireless Hop Count
Without Background Traffic Background Traffic 1.5 MbpsBackground Traffic 3.0 Mbps
Figure 3.4. VoIP performance: Packet-loss rate.
1 2 3 440
60
80
100
120
140
160
180
200
220
Wireless Hop Count
Jitte
r (m
s)
Jitter vs. Wireless Hop Count
Without Background Traffic Background Traffic 1.5 MbpsBackground Traffic 3.0 Mbps
Figure 3.5. VoIP performance: Jitter.
3.4.2.2 Voice-over-IP (VoIP)
Next, we present the VoIP end-to-end performance. We use skype as
the VoIP application. Figures 3.4, 3.5, and 3.6 present different perfor-
68
1 2 3 41
2
3
4
5
Wireless Hop Count
MO
S
MOS vs. Wireless Hop Count
Without Background Traffic Background Traffic 1.5 MbpsBackground Traffic 3.0 Mbps
Figure 3.6. VoIP performance: Mean Opinion Score (MOS).
mance measures for skype-based experiments. As the number of wire-
less hops increases, both packet-loss rate and jitter increase, resulting in
degraded voice quality. Voice quality also degrades with the increase of
background traffic load. We use the performance metric of Mean Opin-
ion Score (MOS) [120] to measure the subjective voice quality. MOS gives
a numerical indication of the perceived voice quality at the receiver end.
MOS is expressed in one number, from 1 to 5, 1 being the worst and 5
being the best. A group of regular VoIP users were asked to give a score
between 1 to 5 after experiencing the voice quality in different experi-
ments. Then, the mean is calculated to determine the MOS for different
experimental setups. By comparing the VoIP performance measures, it
is evident that packet-loss rate increases (hence voice quality (or MOS)
decreases) with the number of wireless hops. As expected, too many
wireless hops will not help to improve the WOBAN performance.
69
1 2 3 40
5
10
15
20
25
30
35
40
Wireless Hop Count
Pac
ket−
Loss
Rat
e (%
)
Packet−Loss Rate vs. Wireless Hop Count
Without Background Traffic Background Traffic 1.5 MbpsBackground Traffic 3.0 Mbps
Figure 3.7. Video streaming performance: Packet-loss rate.
1 2 3 450
100
150
200
250
300
Wireless Hop Count
Jitte
r (m
s)
Jitter vs. Wireless Hop Count
Without Background Traffic Background Traffic 1.5 MbpsBackground Traffic 3.0 Mbps
Figure 3.8. Video streaming performance: Jitter.
3.4.2.3 Video-on-Demand (VoD)
We use Darwin Streaming Server3 as VoD server and VLC Player4 as
client for our video experiments. In this real-time video streaming sce-3“Darwin Streaming Server”, http://developer.apple.com/opensource-/server/
streaming/index.html.
70
(a) (b)
(c) (d)
Figure 3.9. Video streaming performance: (a) Original video, and at1.5 Mbps background traffic video quality (b) After one wireless hop, (c)After two wireless hops, and (d) After three wireless hops.
nario, the VoD server broadcasts the video and the client plays the broad-
casted streaming video. The broadcasted streaming video file is 30 sec.
in duration, 640×480 pixels in size, and encoded at 500 kbps. Figures 3.7
and 3.8 show the corresponding packet-loss rate and jitter, respectively,
with number of wireless hops. Figures 3.9(a)-3.9(d) (screen shots taken
at 17 sec. of the video streaming on the client side) show the qualita-
tive video streaming performance with different number of wireless hops.
In these figures, the background traffic is assumed to be moderate (1.5
Mbps).4“VLC Player”, http://www.videolan.org/vlc/.
71
As the number of wireless hops increases and as expected, the video
packet-loss rate increases, and the video quality deteriorates. Till two
wireless hops, we can receive decent quality of video. After three hops,
the video is blurred (Fig. 3.9(d)), and after four hops only a blank screen
shows up in the video client. A heavily-congested network also signifi-
cantly affects the quality of video transmission. Therefore, the wireless
mesh front-end of the WOBAN should not have many wireless hops if it
has to provide quality broadband services to end users.
3.4.3 Critical Observations
We accumulate the following observations from our WOBAN prototyping
procedure and experiments.
• Many wireless hops do not help. But intelligent Gateway placement
in the wireless mesh may help to reduce the number of wireless
hops, and improve the overall WOBAN performance. We can also
put more Gateways in the mesh to decrease the number of wireless
hops.
• Intelligent channel assignment in the wireless mesh can help to
improve performance. We found that, during our mesh setup, if
channel 1 of the 2.4-GHz band is assigned to the wireless routers,
we can get better results compared to assigning channel 6. This is
due to several other interfering routers (outside of our WOBAN) near
the mesh setup working on channel 6. All the results presented in
this work have been collected using channel 1.
• A dynamic link-state routing protocol such as OLSR wastes a lot of
wireless bandwidth. As the WOBAN front-end is a relatively static
mesh and a small number of wireless hops is needed for improved
performance, the WOBAN mesh performance can be improved by
using static routing.
72
• Wireless nodes near a Gateway carry more traffic compared to dis-
tant ones. Therefore, the memory and processing power of these
“closer” nodes should be higher. Moreover, from prototyping view-
point, current processing power and memory of off-the-shelf wire-
less routers will not be sufficient for virtual slicing (where several
experiments are running on the same physical resources).
• As the optical segment of the WOBAN prototype uses a TDM-based
Medium Access Control (MAC) scheme, for better wireless and op-
tical integration and for improved performance, a TDM-based MAC
would be a better choice for the wireless mesh.
• For video transmission, the standard MAC protocol is not sufficient.
The MAC layer should be able to distinguish and prioritize between
video frames and other traffic for better video performance.
• Although a wireless node can have a theoretical maximum capacity
of 54 Mbps, due to interference and other surrounding interference,
the wireless capacity achieved is much lower.
• Routing in the wireless mesh without considering the optical seg-
ment’s traffic condition does not help, and vice versa. Therefore,
an integrated routing approach will help to improve WOBAN perfor-
mance.
3.5 Research ChallengesIn this section, we discuss some research challenges which we have ac-
cumulated from the experience gathered from our WOBAN prototype de-
velopment.
73
3.5.1 Layer-2 Integrated Routing
Current deployment of WOBAN assumes separate data-transfer tech-
niques for optical and wireless segments. In the optical part, we use
MPCP-based Dynamic Bandwidth Allocation (DBA), whereas the wire-
less mesh uses Layer-3 routing, namely OLSR. So, current WOBAN de-
ployment employs a loosely-integrated network architecture and control.
Layer-3 routing in the wireless mesh also poses significant overhead on
the network. To provide seamless integration of the optical and wireless
segments, and to reduce Layer-3 processing overheads, an interesting
alternative is an integrated Layer-2 (L2) routing protocol which can effi-
ciently route traffic through all segments of WOBAN.
The optical segment of WOBAN already uses MPCP-based DBA, namely
Interleaved Polling with Adaptive Cycle Time (IPACT) [121]. Therefore, one
can develop a hierarchical MPCP-based L2 routing for WOBAN (multi-
point control for an OLT to its downstream ONUs and for an ONU to
its downstream Gateways). The idea of L2 routing can be extended in
the optical segment (till the Gateways) so that it fits with the wireless
mesh architecture with one ONU driving multiple Gateways (similar to
the case where one OLT drives multiple ONUs). The wireless mesh will
use a spanning tree for L2 routing. This approach is consistent with the
idea of end-to-end L2 capability of WOBAN.
3.5.2 TDM MAC for Wireless
Traditional wireless mesh uses collision-based MAC protocols. Our cur-
rent deployment based on IEEE 802.11g wireless routers uses Carrier
Sense Multiple Access with Collision Avoidance (CSMA/CA) MAC proto-
col. From our testbed experience, it is evident that CSMA/CA poses a
hindrance on the limited wireless capacity. From the literature, we find
that a TDM-based MAC protocol can improve the capacity of the wire-
less mesh. Furthermore, as we envisioned for a L2 routing approach
74
earlier, a TDM-based MAC will also be consistent with a L2 routing pro-
tocol. Therefore, a TDM-based MAC protocol for the wireless mesh will
lead to the seamless integration of both optical and wireless segments of
WOBAN. Other MAC protocols like Orthogonal Frequency-Division Multi-
plexing (OFDM) combined with TDM can also be considered in the future
to improve wireless capacity.
3.5.3 Improve Flexibility in WOBAN Architecture
Existing PON technologies do not exhibit sufficient fault tolerance and
self-organization capabilities. In case of OLT, ONU, or wireless gateway
failures in a WOBAN, we need to redirect the traffic to other live nodes.
The self-organization and fault-tolerant properties of WOBAN should en-
sure this flexibility. Moreover, when an ONU gets congested due to heavy
load, we need to perform load shifting and load balancing so that the
network’s health is ensured.
3.5.4 Hierarchical Architecture
From our experimental observations, it is clear that wireless Gateways
and routers near a Gateway carry more traffic compared to routers which
are far away from a Gateway. Therefore, the routers in the vicinity of
the Gateway and the Gateway itself should be well-equipped with high-
capacity wireless resources. The capacity of wireless routers can be in-
creased using technologies such as multiple radios, directional antenna,
Multiple Input Multiple Output (MIMO) systems, etc.
3.5.5 Energy-Efficiency in WOBAN
As discussed in Chapter 2, future WOBAN deployment should incorpo-
rate energy-efficiency features to reduce the environmental impact of
telecom network infrastructures. Intelligent methods to increase net-
work utilization by load-adaptive resource management (developed in
Chapter 4) can be adopted to increase WOBAN’s energy efficiency.
75
3.6 ConclusionIn this chapter, we showed how to build a prototype for a novel, high-
bandwidth future access network technology, named WOBAN. This tech-
nology is envisioned to satisfy future bandwidth demand of technology-
savvy customers in a cost-effective manner, and it can be an attractive
solution for future “last-mile” access networks. We demonstrated the per-
formance of several typical applications such as data transfer, voice, and
video over our WOBAN prototype. We observed that too many wireless
hops degrade the application performance, particularly for video. Future
research challenges accumulated from our prototyping experiences were
also illustrated. The WOBAN prototype will be instrumental to develop,
test, and analyze the performance of hybrid network protocols. This
programmable and configurable access architecture will facilitate future
experimental, hybrid, and cross-domain networking research. Energy-
conservation mechanisms for WOBAN can also be studied and evaluated
by experimentations on this prototype.
76
Chapter 4
Building a Green WOBAN
4.1 IntroductionAccess networks, known as the “last mile” of telecommunication network,
connect the telecom Central Office (CO) to the residential and business
customers. Access network comprises a large part of the Internet. It
is also a major energy consumer in the Internet due to the presence of
huge number of active elements [77]. It is estimated that access network
consumes around 70% of overall telecom network energy consumption,
and it will continue to consume a major portion of overall Internet energy
consumption during the next decade [77]. Hence, effective strategies to
reduce energy consumption in access networks can lead to major savings
in the Internet energy consumption. Access network power-consumption
reduction not only has the potential of significant cost savings, but also
it will allow us to achieve the ultimate goal of developing environment-
friendly technologies to build the future “green” Internet. Therefore, fu-
ture access networks should be green featuring efficient energy manage-
ment schemes to reduce their “carbon footprint.”
There are several access technologies in today’s telecom market –
Digital Subscriber Line (xDSL), Cable Modem (CM), Wireless and Cel-
lular networks, Fiber-To-The-x (FTTx), Wireless-Optical Broadband Ac-
77
cess Network (WOBAN), etc. Since customer demands for bandwidth-
intensive services (such as Video-on-Demand, online Gaming, HDTV,
etc.) are rapidly increasing, future-proof access networks should have
higher capacity. The maximal capacity provided by access technologies
such as xDSL, CM, Wireless, and Cellular networks will soon reach a sat-
uration point in satisfying future Internet demands. FTTx technologies
can provide higher bandwidth but still remain cost-prohibitive. In this
regard, we are witnessing an emergence of hybrid wireless-optical tech-
nologies for multi-gigabit data and video applications [122]. WOBAN –
a novel hybrid access network paradigm with the combination of high-
capacity optical backhaul and wireless front-end – can provide very high
throughput in a cost-effective manner [4], [114].
The cost per byte of traffic in access technologies is reducing over
time, making broadband access affordable to more users. However, there
is significant wastage of electricity (several TWh/year) in the Internet
due to inefficient network and system design [104]. A good portion of
this energy is consumed by idle network elements [103]. An estima-
tion shows that access networking equipment are less than 15% uti-
lized [123]. Therefore, energy can be conserved by reducing the energy
consumption of these idle network elements and increasing network uti-
lization. To reduce the energy consumption of idle network elements,
researchers are developing energy-efficient equipment [79], [103]. There
are two other directions of network energy management – energy-aware
network design and energy-aware protocol design [61]. These two strate-
gies intend to improve network utilization by techniques such as shutting
down under-utilized network elements, energy-aware routing, etc.
In wired access technologies such as xDSL, CM, or FTTX, it is very
hard to shut down under-utilized parts of the network, as this will af-
fect the availability of the network, e.g., shutting down a DSL Access
78
Multiplexer (DSLAM) will leave a large number of end-users unserved.
Same argument holds for regular wireless mesh networks. Combining
wired and wireless access technologies in a hybrid architecture to provide
broadband access, as in WOBAN, not only provides a cost-effective solu-
tion [4], but also enables great opportunities for energy savings. WOBAN
has a wireless front-end which provides flexibility of rerouting traffic to-
wards diverse optical access points, so the network utilization can be im-
proved by shutting down low-load optical elements while rerouting the
affected traffic through other live parts of the network. WOBAN also
allows users to be untethered and mobile, if necessary.
In Chapter 3, we developed a programmable WOBAN prototype which
enables researchers to experiment with WOBAN protocols and algorithms.
In this chapter, we develop energy-aware design techniques and rout-
ing protocols in WOBAN for “green” broadband access. We develop a
mathematical model which will act as a specification of the problem
and as a guideline for energy-aware WOBAN design. To the best of our
knowledge, this is the first work to devise energy-consumption reduction
techniques in hybrid wireless-optical access networks. In future, these
energy-conservation mechanisms for WOBAN can be further investigated
in real networking environments using the WOBAN prototype (developed
in Chapter 3).
The remainder of this chapter is organized as follows. In Section 4.2,
we briefly discuss related work on energy management in access net-
works. Section 4.3 describes the WOBAN architecture, techniques for
energy-aware WOBAN design, and energy-aware WOBAN routing proto-
col. In Section 4.4, we present a case study to illustrate the effectiveness
of our energy-aware design on WOBAN. Section 4.5 includes illustrative
numerical examples on the case study and gives insights on better en-
ergy management. Finally, we conclude in Section 4.6.
79
4.2 Related WorkMany research efforts focus on different aspects of energy management
in telecom networks. The wireless networking community has developed
many techniques for energy-efficient wireless technologies. A survey of
energy-efficient protocols for wireless networks can be found in [124].
Here, we concentrate on efforts to improve the energy efficiency of access
networks.
Several research efforts provide approximate estimations of energy
consumption in different types of access networks [3], [77]. They com-
pare the energy consumption of point-to-point optical links, Passive Op-
tical Network (PON), fiber to the node (FTTN), and WiMAX. The results
show that PON is more energy efficient than point-to-point or active op-
tical access networks.
At system level, different PON technologies (both Ethernet PON (EPON)
and Gigabit PON (GPON)) are being improved for energy efficiency through
efficient IC technologies, better devices, and energy-efficient chips [79].
However, both these PON standards have not incorporated any energy-
efficiency features yet. There are several proposals and recommenda-
tions to improve the energy efficiency of PON variants. There are propos-
als for low-power (sleep) state for PON equipment [80], handshaking pro-
tocol for coordinated sleeping mechanism [84], shedding power in User
Network Interface (UNI) and Access Network Interface (ANI) [79], speed
shedding in UNI and ANI [79], etc. Once incorporated, these techniques
can save energy for both PON standards.
In xDSL, energy efficiency can be improved by reducing electromag-
netic interference, i.e., crosstalk, which happens due to signal interfer-
ence of different lines in the same cable bundle. Crosstalk can hugely
deplete a DSL line’s available bandwidth. Dynamic Spectrum Manage-
ment (DSM) coordinates the spectrum and/or signals from all users to
80
reduce crosstalk [89]. It is estimated that there are opportunities for up
to 50% energy savings while achieving 85% full-power data rate perfor-
mance in real DSL network scenarios [89].
Despite all these efforts, there remain significant challenges to deploy
energy-efficient design in access networks.
4.3 Green WOBANIn this section, we present the architecture of WOBAN, a mathemati-
cal model of energy-aware WOBAN design, and an energy-aware routing
algorithm for WOBAN.
4.3.1 WOBAN Architecture
Hybrid Wireless-Optical Broadband Access Network (WOBAN) is a novel
access network architecture with an optimal combination of an optical
backhaul (e.g., a Passive Optical Network (PON)) and a wireless front-
end (e.g., WiFi and/or WiMAX). Figure 4.1 presents the architecture of
WOBAN. WOBAN optimizes the deployment cost due to less-expensive
wireless front-end and maximizes the bandwidth performance of a broad-
band access network [114].
In WOBAN (Fig. 4.1), a PON segment starts from the Optical Line
Terminal (OLT) at the telecom CO and terminates at multiple Optical
Network Units (ONU). Multiple wireless routers form the front-end of
WOBAN. A selected set of these routers are called gateways. The front-
end of WOBAN is essentially a multi-hop Wireless Mesh Network (WMN)
with several wireless routers and a few gateways. These gateways are
connected to the PON backhaul through the ONUs. Each ONU can sup-
port several wireless gateways. End-users (both mobile and stationary)
connect to WOBAN through the wireless routers.
In WOBAN, when an end-user wants to send a packet, it sends the
packet to its nearest wireless router. The wireless router can deliver the
81
Figure 4.1. WOBAN architecture
packet(s) to any of the gateways. Therefore, in the upstream direction
of the wireless mesh, WOBAN is an anycast network. The gateway can
then send the packet to the ONU connected to it. In the optical back-
haul (from ONUs to OLT), WOBAN is a shared-medium access network
where ONUs contend for the shared upstream channel to OLT in a time
division manner. The optical backhaul in the downstream is a broadcast
network where packets are broadcasted to all the ONUs. Only the des-
tination ONU keeps the packet, while others discard them. However, in
the downstream direction from the wireless gateways, WOBAN is a uni-
cast network, since a gateway will send the downstream packets towards
the specified destination routers.
4.3.2 Energy-Aware WOBAN Design
WOBAN represents a hierarchical access architecture with gateways as
the initial traffic aggregation points. ONUs are the next aggregator level
in the hierarchy, while OLT is the highest aggregation point and connects
the access network with the rest of the Internet.
82
4.3.2.1 Enabling Factors for Energy Savings in WOBAN
Several aspects of WOBAN need to be considered for its energy-aware de-
sign. Current WOBAN design, deployment, and management methods
provide fault tolerance, reliability, and robustness as well as a high level
of performance. Thanks to the mesh front-end, traffic can be rerouted
through alternate paths in case of failures such as a fiber cut, or a fail-
ure of a wireless router or gateway or ONU. Moreover, there is a capac-
ity mismatch between the wireless front-end and the optical backhaul.
The redundant capacity in the optical backhaul can provide extra reli-
ability during the failure so that traffic can be rerouted through alter-
nate paths1. At a specific instant, it is possible to find several WOBAN
topologies that can satisfy the required capacity and reliability objec-
tives. All these are possible due to the densely-interconnected wireless
mesh front-end which has many redundant paths to route traffic. The
flexibility provided by the wireless front-end of WOBAN can be exploited
to enable energy savings in the optical part.
Another important aspect of the access network is its traffic profile.
The traffic load on the access network comes directly from customers,
and it is well known that there are daily fluctuations of this load. During
WOBAN (as well as other network) deployment, the common practice is
to deploy network equipment so that they can support the peak traffic
load. Consequently, during low-load hours, some parts of the network
may be under-utilized.
Hence, to design WOBAN topologies with reduced power consump-
tion, we need to consider the following points – (a) a WOBAN topology can
provide several redundant paths for a packet to reach its destination, and
(b) traffic load variation during different hours of the day. Thus, we can1Note that, in general, the high capacity of the PON can not only serve as the back-
haul of the wireless front-end but also serve other wired business and residential cus-tomers (but this traffic is not considered here).
83
selectively put some nodes to a low-power (sleep) state during low-load
hours, thereby reducing network-wide power consumption.
In the wireless front-end of WOBAN, we need to keep all the wireless
routers on to ensure availability of the network. In this work, we mainly
focus on how to put optical components of WOBAN into sleep state. We
will not consider putting OLT into sleep state as it connects the WOBAN
to the rest of the Internet. However, for protection purposes, in a PON
segment, it is possible to have several OLTs in a ring setup. In that
case, a low-load OLT can be put into sleep state while rerouting its traffic
through other OLTs. In this work, we reduce ONU power consumption
in WOBAN by putting low-load ONUs to sleep.
Algorithm 1 Coordinated ONU Shut-Down AlgorithmInput: WOBAN topology, Low Watermark (LW), and High Watermark
(HW).
Output: Set of ONUs that can be shut down.
• Initialization: Initialize LW and HW.
• Measurement: At different hours of the day, OLT quantifies traffic
load at different ONUs by measuring the length of corresponding
input queues (maintained by the OLT).
• Decision: ∀ ONUs,
– If load < LW, shut down ONU.
– else if load > HW, keep ONU active and turn on another inac-
tive ONU.
– else keep ONU active.
84
4.3.2.2 How to Put an ONU to Sleep State
Current IEEE 802.3ah/ 802.3av standards do not define any low-power
state for an ONU [125]. However, proposals have been made to IEEE
802.3av task force to include low-power states for an ONU so that it can
go to sleep during idle periods [80]. Typical power consumption by an
ONU during active state is approximately 10 W [126]. It is also estimated
that, during sleep state, power consumed by an ONU is less than 1 W
[80]. Existing ONUs in the market include a TX_DISABLE input which
disables the transceiver of an ONU [126]. Disabling the transceiver can
reduce ONU power consumption several fold.
In WOBAN, the OLT can manage a centralized sleeping mechanism to
put low-load ONUs to sleep. The mechanism works as given in Algorithm
1. An OLT maintains two watermarks for the traffic load at ONUs – low
and high watermark.
The wireless mesh front-end of WOBAN will reroute to alternate paths
the affected traffic due to ONU shutdown.
4.3.2.3 Mathematical Model
Now, we determine the optimal number of ONUs needed to support a
given amount of traffic load. This is a multi-commodity flow problem
where each commodity represents the traffic flow between a source-
destination pair. We formulate the problem as a Mixed Integer Linear
Program (MILP).
Our model takes as input a WOBAN with preassigned link capacities
and traffic loads (i.e., traffic matrix) based on the dynamic daily traffic
profile of a service area. The model generates output which determines
minimum number of ONUs that need to be kept active to route the given
traffic load. The other ONUs can be put to sleep so that network-wide en-
ergy consumption is minimized. The model also finds the routing path
for each source-destination (s, d) pair’s traffic flow. In WOBAN, for up-
85
stream traffic, the destination is always the OLT while the source can be
one of the wireless routers, and for downstream traffic, the source is the
OLT and the destination is one of the wireless routers.
To describe the model, we introduce some notations for the parame-
ters and variables as follows.
• WOBAN topology: denoted by a weighted and directed graph G =
{V,E} where V is the set of nodes and E is the set of links. V has
three subsets – Vw is the set of wireless nodes, Vonu is the set of
ONUs, and OLT represents the OLT. If nodes u and v have a link,
the link is denoted by (u, v). E has several subsets – Ew is the set of
wireless links, EOG is the set of ONU-to-Gateway links, EGO is the
set of Gateway-to-ONU links, ETO is the set of OLT-to-ONU links,
and EOT is the set of ONU-to-OLT links.
• (s, d): Identifies source-destination pair of the upstream/downstream
traffic in the traffic matrix.
• Xu: Binary variable denoting ONU state, Xu ∈ {0, 1}. 0 denotes ONU
is asleep, and 1 denotes ONU is active.
• λs,du,v: Binary variable denoting downstream flow on link (u, v) for a
(s, d) (s is OLT, d denotes routers) pair, λs,du,v ∈ {0, 1}.
• γs,du,v: Binary variable denoting upstream flow on link (u, v) for a (s, d)
(s denotes routers and d is OLT) pair, γs,du,v ∈ {0, 1}.
• Cu,v: Variable expressing the capacity over a wireless link (u, v). This
variable can assume non-integral values.
• COG: Capacity of ONU-to-Gateway link.
• CGO: Capacity of Gateway-to-ONU link.
• COT : Capacity of ONU-to-OLT link.
86
• CTO: Capacity of OLT-to-ONU link.
• T : Input traffic matrix with two different types of traffic values –
(1) Vs,d: Downstream traffic between a (s, d) pair and (2) Zs,d: Up-
stream traffic between a (s, d) pair. For each (s, d) pair, we assume
the upstream traffic to be a fraction of the downstream traffic, i.e.,
Zs,d =Vd,s
ft, ∀s,d where ft is a constant value.
Now, the objective function can be written as:
minimize∑
u∈Vonu
Xu (4.1)
subject to the following constraints:
Flow Constraints: Equation (4.2) captures the fact that, in all nodes
of WOBAN, total outgoing downstream traffic should be equal to total
incoming downstream traffic except for the source (OLT) and the desti-
nation nodes (wireless routers). Similar argument holds for upstream
traffic (Eqn. (4.3)) except for the source nodes (wireless routers) and the
destination (OLT).
∑(u,v) Vs,dλ
s,du,v −
∑(v,u) Vs,dλ
s,dv,u =
−Vs,d, u = d
+Vs,d, u = OLT
0, otherwise
∀u ∈ V,∀(s, d) ∈ T
(4.2)
∑(u,v) Zs,dγ
s,du,v −
∑(v,u) Zs,dγ
s,dv,u =
+Zs,d, u = s
−Zs,d, u = OLT
0, otherwise
∀u ∈ V,∀(s, d) ∈ T
(4.3)
Wireless Capacity Constraint: A wireless link in WOBAN can carry
both upstream and downstream traffic. Equation (4.4) states that the
summation of all traffic through a wireless link (u, v) should not exceed
the capacity (Cu,v) of the link.
87
∑(s,d)
(Vs,dλs,du,v + Zs,dγ
s,du,v) ≤ Cu,v, ∀(u, v) ∈ Ew (4.4)
Wireless Constraints: The dual-threshold interference model [127] is
used to find the set of all interfering links at each wireless node of WOBAN.
The wireless interference constraints are translated to constraints that
allocate capacity on each wireless link. A wireless node divides its ca-
pacity to all its incoming and outgoing links as it can not transmit and
receive at the same time. So, the interference-free radio capacity avail-
able (Cu) at each node u is shared between all the outgoing links from u
(first term in Eqn. (4.5)) and all the incoming links to u (second term in
Eqn. (4.5)). ∑v
Cu,v +∑v
Cv,u ≤ Cu, ∀u ∈ Vw (4.5)
Equation (4.6) forms the secondary interference constraint of the wire-
less mesh of WOBAN. This constraint states that a node cannot receive
any signal from any other node when an interfering link is active. The
first term of Eqn. (4.6) is the same as in Eqn. (4.5), representing the
shared capacity among incoming links to node u. The second term rep-
resents all the links which interfere with node u (Iu,v) and which do not
have node u as one of their end points.∑v
Cv,u +∑
(p,q)∈Iu,v
Cp,q ≤ Cu, ∀u ∈ Vw (4.6)
Wired Capacity Constraints: Downstream traffic flows are limited by
the capacity of the ONU-to-GW (Eqn. (4.7)) and OLT-to-ONU (Eqn. (4.8))
links. Similarly, upstream traffic flows are limited by the capacity of the
GW-to-ONU (Eqn. (4.9)) and ONU-to-OLT (Eqn. (4.10)) links.∑(s,d)
Vs,dλs,du,v ≤ COG, ∀(u, v) ∈ EOG (4.7)
∑(s,d)
Vs,dλs,du,v ≤ CTO, ∀(u, v) ∈ ETO (4.8)
88
∑(s,d)
Zs,dγs,du,v ≤ CGO, ∀(u, v) ∈ EGO (4.9)
∑(s,d)
Zs,dγs,du,v ≤ COT , ∀(u, v) ∈ EOT (4.10)
Wired Directionality Constraints: These constraints ensure that no
upstream traffic is flowing in the downstream direction in the wired part
of WOBAN and vice versa.
λs,du,v = 0, ∀(u, v) ∈ EOT ∪ EGO, ∀(s, d) ∈ T (4.11)
γs,du,v = 0, ∀(u, v) ∈ ETO ∪ EOG, ∀(s, d) ∈ T (4.12)
ONU State Constraint: This constraint determines the ONU state. If
some traffic (upstream (γs,du,v) or downstream (λs,d
u,v)) flows through an ONU
u, it should be active (Xu = 1), otherwise it should be in sleep state (Xu =
0). This condition can be expressed by the following constraint:
Xu ≥∑
v
∑(s,d) λ
s,du,v +
∑v
∑(s,d) γ
s,du,v
M, ∀u ∈ Vonu (4.13)
where M is a very large value used to map (i.e., to normalize) the flow
variables into a binary variable (Xu).
Path-Length Constraints: These constraints put a limit on the path
length. As Eqns. (4.14) and (4.15) show, each upstream or downstream
(s, d) path should not be longer than H hops.∑u,v
λs,du,v ≤ H, ∀(s, d) ∈ T (4.14)
∑u,v
γs,du,v ≤ H, ∀(s, d) ∈ T (4.15)
This formulation turns out to be a MILP as some variables (such as
Cu,v) can take non-integral values. This specific MILP formulation is NP-
hard due to large number of variables and constraints for a network with
89
large number of nodes. However, sophisticated heuristics can be used
to solve it in a reasonable amount of time.
We try to solve our model with smaller networks to verify the correct-
ness of our problem formulation and to provide a performance bench-
mark for our heuristics. The MILP and heuristics will be compared later.
4.3.3 Energy-Aware Routing
For a WOBAN with a larger number of nodes and high traffic load, we
need to build some heuristics to solve this problem. Our first heuristic
decides which ONUs to shut down, i.e., “put ONUs with load less than
low watermark to sleep” (see Algorithm 1). Our second heuristic develops
an energy-aware routing method as discussed below (see Algorithm 2).
Several routing protocols have been proposed for WOBAN-like archi-
tectures, e.g., Delay Aware Routing Algorithm (DARA) [4] and Capacity
and Delay Aware Routing (CaDAR) [4]. These algorithms are Link-State
(LS) protocols where a node periodically transmits its link-state informa-
tion to the network by Link-State Advertisement (LSA). Upon receiving
the LSAs from all the nodes, each node finds a map of the network and
can build a routing table (generally by using some variant of Dijkstra’s
algorithm) to route traffic to other nodes in the network. LS protocols
generally vary on how they assign link weights in the LSA. For exam-
ple, DARA uses predicted link delay metric to assign link weights. Based
on link weight assignment, these protocols try to achieve several perfor-
mance objectives. One such objective is load balancing which balances
the traffic load in all parts of the network [4].
Load balancing is a good performance objective as it tries to fairly
utilize all parts of the network. But it may lead to under-utilization of
some segments of the network during low-load hours. During low-load
hours, traffic can be supported using a small number of devices in the
network. Our routing algorithm is an energy-aware LS protocol whose
90
objective is to reduce the network-wide energy consumption by putting
under-utilized nodes (mainly ONUs) of the network to sleep. When rout-
ing traffic, the objective will be to “use the already-used paths.” Thus,
zero-load ONUs can be put to sleep. Moreover, we may find some other
ONUs with very low load (ONUs with loads under low watermark). By be-
ing more aggressive, we can put these ONUs to sleep and let the wireless
mesh reroute their traffic through other active ONUs. When traffic load
increases, sleeping ONUs can be activated to carry the increased traffic.
To achieve this, we can modify the LS routing algorithm in WOBAN so
E
C B
H
G F
D S
3
2
7 6
3
3
1
2
2
Figure 4.2. Residual capacity as link weights.
that link weights are assigned to satisfy our energy objective. So, we
use residual capacity as the link weight. When traffic flows through a
link, its next link weight will be the remaining capacity (original capacity
minus traffic flow) on that link. To route traffic from source to desti-
nation, we find the lowest residual capacity path. A formal description
of the various steps of our algorithm can be found in Algorithm 2. For
example, let us consider the small network in Fig. 4.2. The link metrics
are their residual capacities. To send traffic from S to D, energy-aware
routing algorithm will route traffic through the path S-E-F-G-D which
has the lowest residual capacity (2 + 2 + 2 + 1 = 7).
This approach, however, has its shortcomings as shown in Fig. 4.2.
91
Algorithm 2 Energy-Aware Routing Algorithm
• Initialization
– For each link (u, v), assign initial capacity as residual capacity.
• Link State Advertisement (LSA)
– For each link (u, v) from node u, advertise periodically the resid-
ual capacity Cuv as link weight and time stamp to other nodes.
• Link Weight Assignment
– For each link (u, v), update the link weight found from the LSA
by adding Hop Offset (HO).
• Path Computation
– Find the lowest residual capacity path between source and des-
tination.
– Update residual capacity of links on the selected path.
The algorithm selects the path with 4 hops although that is not the short-
est path while using other metrics (such as hop length or delay). This
will increase the average path length and path delay in the network. To
deal with this problem, we can introduce a term called hop offset – the
purpose of this term is to reduce average path length. If we have a hop
offset m, we add m to the path cost for each hop, i.e., for a path of n hops,
the cost of the path will be residual capacity of the path + n × m. For
example, as in Fig. 4.2, for a hop offset 1, the path costs are 7+4×1 = 11,
9+3× 1 = 12, and 13+2× 1 = 15 for paths S-E-F-G-D, S-B-C-D, and S-H-
D respectively, and S-E-F-G-D will be the selected path. But, for a hop
offset 3, S-B-C-D will be the chosen path in our algorithm, and for a hop
offset 5, S-H-D will be the chosen path. Selecting the optimal hop offset
92
depends on how much delay the network connections can tolerate.
There is another important item to consider. We should select hop
offset in such a way that average path length does not increase unpro-
portionately from regular shortest-path routing. If average path length
increases too much, that means more wireless hops per path, i.e., more
wireless transmissions and receptions. Each wireless transmission/rece-
ption requires power. So, out-of-proportion average path length may di-
minish the power savings that we gain from putting ONUs to sleep. We
do not elaborate on systematically setting the routing metric and LS, as
this energy-efficient routing can be adopted in any LS routing algorithm
with residual capacity as link weight.
4.4 Case StudyFigure 4.3 shows a hypothetical WOBAN deployment scenario in Davis,
which is a small city in Northern California near Sacramento. Davis is
the home of the University of California, Davis. The selected part of Davis
has three different areas: (a) Downtown, (b) UC Davis Campus, and (c)
Part of residential area. These areas are selected as they have a very nice
blend of technology-savvy users, and we can showcase how traffic profile
varies across different parts of the network, and also depending on the
traffic profile during different time of the day, how we can put nodes to
sleep.
The telecom CO is situated in the downtown area. Total 140 wireless
routers, 24 gateways, 12 ONUs, and 1 OLT are deployed in this hypothet-
ical deployment. The OLT drives 12 ONUs, and 1 ONU drives 2 gateways.
Downtown has 41 routers, 8 gateways, and 4 ONUs; Campus area has
37 routers, 6 gateways, and 3 ONUs; and Residence area has 62 routers,
10 gateways, and 5 ONUs deployed. The routers are equipped with one
radio with a capacity of 54 Mbps (IEEE 802.11g). Average distance be-
93
Telecom CO
OLT
Splitter
ONU
Wireless Gateway
Wireless Router
Legends:
Map Courtesy: Google Maps
Optical Fiber
CAT-5 Cable
(b) UC Davis Campus (a) Davis Downtown
(c) Part of Davis Residential Area
Figure 4.3. Hypothetical WOBAN deployment in Davis.
tween wireless routers is 50 m. Capacity allocation among wireless links
is accomplished by TDM link scheduling. The OLT and ONU have capac-
ities of 1 Gbps and 100 Mbps, respectively. The low watermark is set to
5%, i.e., the OLT puts ONUs with load less than 5% of the total traffic to
sleep. The high watermark is set to 80%, i.e., when an ONU has more
than 80% traffic load, we need to activate another inactive ONU to carry
the extra traffic. Selecting different watermark values has significant
impacts on the performance which will be shown later. We consider that
94
the time needed to shut down an ONU and to bring it up are included in
the ONU’s sleeping duration.
4.4.1 Traffic Modelling
For our illustrative examples, we need to develop reasonable traffic pro-
files of the deployment areas during different hours of the day. Access
networks deal directly with the user-generated Internet traffic. So, the
behavior of end-users has a significant impact on the performance of the
access network. By following the access network traffic models in [128],
[129], [130], we develop a traffic profile for each of the three different
areas of Davis. Each of these areas has different types of Internet users
with different behavior patterns and different peak usage periods. It is
worth noting that our study is generic and can take any traffic model as
input.
1 2 3 4 5 6 7 80
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Hours of the Day
Per
cent
age
of A
ctiv
e R
oute
rs
Downtown AreaCampus AreaResidence Area
00−03 03−06 06−09 09−12 12−15 15−18 18−21 21−24
Figure 4.4. Traffic profile: Ratio of active routers.
We divide the 24 hours of a day into 8 periods: Period 1 - (00-03 hours),
Period 2 - (03-06 hours), Period 3 - (06-09 hours), Period 4 - (09-12 hours),
95
1 2 3 4 5 6 7 80
0.5
1
1.5
2
2.5
3
Hours of the Day
Ave
rage
Loa
d at
Act
ive
Rou
ters
(M
bps)
Downtown AreaCampus AreaResidence Area
00−03 03−06 06−09 09−12 12−15 15−18 18−21 21−24
Figure 4.5. Traffic profile: Average load on active routers.
Period 5 - (12-15 hours), Period 6 - (15-18 hours), Period 7 - (18-21 hours),
and Period 8 - (21-24 hours). Period 1 begins at 00.01 AM in the morning.
The granularity of these periods can be modified as necessary. Figures
4.4 and 4.5 show such a traffic profile for downtown, campus, and res-
idential areas of Davis. As presented in [129], an access network traffic
profile can be modelled by two components - (a) how many end devices (in
our case wireless routers) are active and (b) what is the average load on
those active devices. In our traffic profile, we also have two parts – Fig.
4.4 presents the percentage of active routers (which is directly propor-
tional to the active users) in these areas, and Fig. 4.5 shows the average
(Poisson-distributed) load on these active routers. Both of these data
together generate the traffic profiles of these areas. We assume that, for
each active user, the upload traffic is approximately 14
of the download
traffic.
96
00−03 03−06 06−09 09−12 12−15 15−18 18−21 21−240
10
20
30
40
50
60
Hours of the Day
% o
f O
NU
s Sh
ut D
own
Regular ModePower−Save ModeEnergy−Efficient Routing
Figure 4.6. Power savings in energy-aware WOBAN.
00−03 03−06 06−09 09−12 12−15 15−18 18−21 21−240
10
20
30
40
50
60
70
Hours of the Day
Wat
ts
ONU Power Savings (Power−Save Mode)Extra Power Consumption in Wireless (Power−Save Mode)ONU Power Savings (Energy−Efficient Routing)Extra Power Consumption in Wireless (Energy−Efficient Routing)
Figure 4.7. Power savings vs. extra wireless power.
4.5 Illustrative Numerical ExamplesWe apply our energy-aware design and routing protocol on the hypo-
thetical WOBAN deployment in Davis (Fig. 4.3). We collect results for
97
three different setups: (a) WOBAN in regular mode with no energy-saving
techniques, (b) WOBAN in power-save mode (with regular shortest-path
routing) where ONUs under low watermark are put to sleep (Algorithm
1), and (c) WOBAN in energy-aware routing mode where energy-aware
routing is deployed on top of power-save mode configurations (Algorithm
2 on top of Algorithm 1). We compare the energy savings and measure
the impact of ONU shutdown on the performance of the network.
We quantify power savings in terms of percentage of ONUs in sleep
state during a certain period. Figure 4.6 shows the power savings during
different periods of the day. Obviously, there is no energy savings in the
regular mode, hence no data for regular mode in Fig. 4.6. Interestingly,
on an average, we can put 50% of the ONUs to sleep state in this scenario
by using the other two setups. One may argue that if we can put 50% of
the ONUs to sleep, why do we deploy them in the first place? The answer
is that, at high load, when all the routers are active, we will not be able
to put any ONU to sleep. The power-saving opportunity lies somewhere
else – specifically in the traffic profile. When one part of the network is
at high load, the other parts may be in low load. We can save energy by
putting ONUs in those low-load parts to sleep.
In the wireless front-end of WOBAN, wireless nodes when inactive can
enter idle state, saving some energy. Existing wireless routers can save
up to 30% less power in idle state than in active (transmitting/receiving)
state [131]. Now, if we shut down ONUs using our energy-saving meth-
ods, we need to reroute some traffic through idle (idle during regular
mode) wireless routers, incurring more wireless energy consumption.
We quantify how much excess energy in wireless routers is consumed
due to rerouting. Figure 4.7 shows that the excess wireless energy con-
sumption is still very less (at given WOBAN configuration) compared to
energy savings by shutting down ONUs in the optical part of WOBAN.
98
00−03 03−06 06−09 09−12 12−15 15−18 18−21 21−240
0.5
1
1.5
2
2.5
3
3.5
4
Hours of the Day
Ave
rage
Pat
h L
engt
h
Regular ModePower−Save ModeEnergy−Efficient Routing
Figure 4.8. Energy-aware WOBAN performance: Average path length.
00−03 03−06 06−09 09−12 12−15 15−18 18−21 21−240
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Hours of the Day
Ave
rage
Pat
h D
elay
(Se
c)
Regular ModePower−Save ModeEnergy−Efficient Routing
Figure 4.9. Energy-aware WOBAN performance: Average path delay.
99
00−03 03−06 06−09 09−12 12−15 15−18 18−21 21−240
10
20
30
40
50
60
Hour of the Day
ON
U U
tiliz
atio
n (%
)
Regular ModePower−Save ModeEnergy−Efficient Routing
Figure 4.10. Energy-aware WOBAN performance: ONU utilization.
Figure 4.8 presents the average path length in three different se-
tups of WOBAN. In WOBAN, all (s, d) paths have two wired (OLT⇔ONU,
ONU⇔GW) hops, and the rest are wireless hops. The average path lengths
in the energy-aware routing mode are comparable with the regular mode.
Hop Offset tries to reduce the average path length in energy-aware rout-
ing mode. Otherwise, higher average path length could diminish our
power conserving benefits with the extra wireless transmission/reception
power consumed. We can see that putting ONUs to sleep does not sig-
nificantly increase the average number of wireless hops, thanks to the
availability of various similar-cost paths in the wireless mesh of WOBAN.
Hence, the energy usage in the wireless part does not increase signifi-
cantly when we put ONUs to sleep.
Figure 4.9 provides the average transmission delay of the paths for
different setups. Again, the transmission delays are very much com-
parable in all these setups. Figure 4.10 shows average ONU utilization
in different scenarios. As mentioned before, access network elements
100
are very under-utilized, leading to extra energy consumption. We can
see that, by using energy-saving mechanisms, we can significantly im-
prove (in some cases, more than 50%) the utilization of the ONUs and
improve utilization-to-energy-consumption ratio. Consequently, we can
state that we can save a good portion of energy consumption in WOBAN
by careful design and energy-aware routing without compromising the
performance.
Table 4.1. Energy savings vs. Low watermark.
Low Watermark (%)% of ONUs Shut Down
Regular
Mode
Power-Save
Mode
Energy-Efficient
Routing Mode
0 0 7.29 46.88
1 0 11.46 47.92
5 0 37.50 54.17
10 0 48.96 63.54
Next, we study the impact of different low watermarks on the perfor-
mance of energy-aware WOBAN. Table 4.1 gives a comprehensive insight
on this. These results are showing averages of 24 hours of a day. It is
obvious that, by increasing the low watermark, we can aggressively put
more ONUs to sleep state. But it is worth noting that, even for lower val-
ues of low watermark, energy-aware routing provides more scope on en-
ergy savings. The reason is that it routes the traffic through the “already-
used" ONUs, leaving some other ONUs unutilized. However, increasing
the low watermark increases the average path length and path delay. Up
to 5% low watermark, average path length and path delay are similar in
regular and power-save mode, and in energy-efficient routing mode, av-
erage path length and path delay increase slightly. However, our results
show that, at 10% low watermark, the impact is very high for both power-
101
save mode and energy-efficient routing mode [(average path length, path
delay) - (3.02, 0.385 sec) (regular), (4.05, 0.626 sec) (power-save), and
(4.6, 0.858 sec) (energy-efficient routing)]. The impact of energy savings
on various aspects of the network’s performance can be explored in fu-
ture. Therefore, we need to be careful in selecting the low watermark.
From our results, we can see that selecting the low watermark at 5%
gives us better scope for energy savings as well as less impact on net-
work performance. These results also show another interesting factor in
energy savings. If the Service Level Agreement (SLA) between end-users
and network operator can offer more tolerance on network-wide delay,
the scope of energy savings can be increased by choosing a higher low
watermark. The effect of different high watermark values is intuitively
similar.
4.5.1 MILP vs. Heuristics
So far, we have presented results from our heuristic methods. To validate
the effectiveness of the heuristics, we compare the results obtained from
the MILP and the heuristic method with Algorithms 1 and 2 combined.
The MILP is not able to produce optimization results for a large WOBAN
(such as the one in Fig. 4.3) in a reasonable amount of time. We use a
moderate 43-node WOBAN with 30 wireless routers, 8 gateways, 4 ONUs,
and 1 OLT for this comparison. Each ONU can drive 2 gateways. The
traffic profile and node distribution in areas are similar to the ones in
Sec. 4.4. In this network, the MILP can be solved in a reasonable amount
of time. For solving the MILP, we use ILOG CPLEX [132] on a Intel Core
2 Duo machine with 1 Gigabyte RAM and Ubuntu Linux OS. For fair
comparison, in the MILP, there is no limit on average path length (Eqns.
(4.14) and (4.15)) and there is no low watermark in the heuristic method,
so only ONUs not carrying any traffic will be shut down since the MILP
does the same.
102
00−03 03−06 06−09 09−12 12−15 15−18 18−21 21−240
0.5
1
1.5
2
2.5
3
3.5
4
4.5
Hours of the Day
Ave
rage
Pat
h L
engt
h
No Power SavingMILPHeuristics
Figure 4.11. Performance of MILP and heuristics: Average path length.
The heuristic method closely follows the MILP results in energy sav-
ings. The reason is, similar to MILP, the heuristic method also tries
to minimize energy consumption by routing traffic through the already-
used paths. As the MILP model tries to maximize ONU energy savings,
average path length obtained by MILP is higher (since there is no limit
on average path length in the MILP) than no-power-saving mode (where
all ONUs are on) with shortest-path routing (Fig. 4.11). We can reduce
this path length by setting H in Eqns. (4.14) and (4.15) to a reasonable
value. The heuristic method provides slightly higher average path length
compared to MILP. Hence, we can deploy these heuristics to obtain near-
optimal performance in an energy-aware WOBAN.
4.6 ConclusionIn this work, we showed how energy consumption for providing broad-
band access using a hybrid wired-wireless access architecture (WOBAN)
can be efficiently reduced. We developed a model which can act as
103
a guideline on designing an energy-aware WOBAN. We devised several
techniques for an energy-aware WOBAN which can effectively reduce the
energy consumption. We analyzed the impact of these energy-aware de-
sign decisions on the performance of the network. With suitable design
parameters, we can achieve comparable performance (of WOBAN) with
the energy-aware mode. The energy savings in the optical part of WOBAN
also does not significantly increase the energy usage in the wireless part.
These energy-aware design techniques applied on WOBAN can be gener-
alized so that they are also be applicable to other access networks such
as PON variants. Future work on this topic may include - (1) study-
ing the performance of green WOBAN with detailed analysis of packet
loss rate, jitter, delay vs. hop length, hop length vs. energy cost, etc.,
(2) energy-aware routing algorithm considering the wireless properties
such as channel condition, modulation method, etc., (3) experimental
evaluation of green WOBAN using WOBAN prototype.
104
Chapter 5
Energy-EfficientMixed-Line-Rate (MLR) NetworkDesign
5.1 IntroductionInternet traffic is continuing to grow overwhelmingly, and the energy us-
age of Internet infrastructures and devices is also growing rapidly. It is
estimated that power consumption of the Internet is around 4% of the
total energy consumption in broadband-enabled countries, and back-
bone network infrastructures (i.e., routers, transmission systems, opti-
cal switches, ROADMs, etc.) consume approximately 12% of the total
Internet energy usage (and is to increase to 20% by 2020) [133]. Carbon
footprint of the Internet is dominated by its energy consumption, so an
obvious way to reduce the carbon emission is to design energy-efficient
network infrastructures (Chapter 2).
As the Internet continues of grow, traffic demands in the Internet
are becoming more heterogeneous. Existing optical backbone networks
support 10-40G line rate, and demands for higher bandwidth are grow-
ing. Recently, a major social networking site claimed that it could use
100G line rate right now if available [134]. Then, why not migrate to an
105
all-100G backbone network? Although 100G deployment increases the
network capacity, it also increases network cost due to the requirement
of expensive transponders [135]. Moreover, due to heterogeneous traf-
fic demands, some parts of a backbone network may not require 100G
capacity. Hence, future cost-efficient optical backbone networks will be
required to support mixed line rates (MLR) (e.g., 10/40/100G). MLR net-
works provide versatility in provisioning bandwidth demands since low-
data-rate requests can be multiplexed into high-capacity wavelengths,
and direct lightpath can be set up on high-capacity wavelengths for high-
data-rate requests [135]. Legacy wavelength-division-multiplexed (WDM)
systems were typically carrying 10G channel rates. On those systems,
40G channel rates have already been introduced and 100G deployment
is imminent. Ideally, these different line rates should coexist over legacy
channel grids and transmission systems. Therefore, a next-generation
WDM system needs to be versatile to support mixed line rates.
In such an MLR network, different wavelengths on a link may carry
different line rates [135], [136]. However, co-propagating wavelengths
with different line rates induce non-linear effects (e.g., cross phase mod-
ulation (XPM), etc.) on each other and reduce the transmission reach
(maximum propagation distance without signal regeneration) of different
line rates [137] based on an acceptable bit-error rate (BER) threshold.
An ideal MLR network should ensure the maximum reach of each line-
rate signal even when it is co-propagating with other line-rate signals. It
has been shown that non-linear effects can be reduced with dispersion
management, channel plan, modulation formats, channel input power
management, etc. [137], [138]. Still, the effect of co-propagating wave-
lengths (at various line rates) on each other’s reach can not be totally
nullified. Therefore, in designing future MLR networks, we should con-
sider whether the mixing of line rates has any detrimental effect on the
106
reach and the capacity of the network.
Then comes the energy consumption of the network. High-data-rate
transmission also consumes more energy compared to low-data-rate tran-
smissions. Hence, in an MLR network, a tradeoff exists between capacity
and energy consumption. High-data-rate wavelengths increase the ca-
pacity and energy consumption of the network at the same time. There-
fore, while designing an MLR network, we need to find the optimum num-
ber of wavelengths at different data rates to support a given set of traffic
demands and minimize the networkwide energy consumption.
Recent network deployment trends suggest that, in backbone net-
works, IP over Wavelength Division Multiplexing (WDM) technologies are
becoming more dominant. In such an IP-over-WDM (IoWDM) network, IP
datagrams are directly carried over optical wavelength channels, thereby
reducing the overhead incurred by in-between electronic layers. In this
work, we develop mathematical models to design energy-efficient MLR IP-
over-WDM (IoWDM) networks. We compare the energy consumption of
both MLR and SLR (Single-Line-Rate) IoWDM networks using our mod-
els. Our results indicate that an MLR network performs better than the
SLR networks by reducing the networkwide energy consumption. From
now on, when we use the words “energy cost”, we mean the energy con-
sumption of networks/elements.
The rest of this chapter is organized as follows. Section 5.2 briefly de-
scribes three different IoWDM architectures. In Section 5.3, we present
the models to design energy-efficient MLR IoWDM networks. Section 5.4
present illustrative numerical examples and relevant discussions. Fi-
nally, concluding remarks are provided in Section 5.5.
107
5.2 IP-over-WDM Network ArchitecturesIn an IP-over-WDM (IoWDM) network, the IP layer is supported by the
underlying optical layer. In the optical layer, the quality of an optical sig-
nal degrades as it travels through the physical medium (e.g., fiber and
other optical components) due to chromatic and polarization-mode dis-
persion, non-linear effects, noise accumulated from optical amplifiers,
etc. To increase the reach of optical signal, the signal should be regener-
ated, amplified, and reshaped at the intermediate locations of the opti-
cal lightpath (unidirectional point-to-point connection). The optical sig-
nal can either be regenerated through optical-electronic-optical (O-E-O)
conversion or in the optical domain. To ensure error-free delivery of bit-
streams over optical channels by reducing the above-mentioned optical
impairments, several IoWDM network architectures have been proposed
over the past few years - some of them are standardized and deployed in
operational networks. In this work, we present three prominent IoWDM
network architectures, namely Transparent, Translucent, and Opaque
architectures, and propose methods to design energy-efficient MLR net-
works based on these architectures. Below, we elaborate on these three
different IoWDM network architectures.
5.2.1 Transparent Architecture
Ideally, in a transparent IoWDM network, optical channels carry IP data-
grams irrespective of bit rates and formats [139]. A transparent opti-
cal WDM network allows optical signals to bypass intermediate nodes,
thereby reducing extra electronic signal processing [140], [141]. In such
a network, O-E-O conversion for signal regeneration does not occur at
intermediate nodes of a lightpath. As there is no O-E-O conversion, the
geographical reach of a transparent network will also be limited as max-
imum transmission reaches of optical signals are limited due to signal
108
Figure 5.1. Transparent IoWDM architecture.
impairments. However, using traffic grooming techniques, several light-
paths can be concatenated to transfer data over longer distances in a
transparent network. In such a case, the intermediate grooming nodes
will incur electronic processing costs, and signals will be regenerated.
As shown in Fig. 5.1, each node in a transparent IoWDM network is
equipped with an optical crossconnect (OXC) attached to an IP router.
In this architecture, data can be transferred in two ways: (a) through
grooming in the electronic layer by IP routers and (b) through direct light-
path by bypassing intermediate IP routers. In Fig. 5.1, Req2 has been
groomed on an existing lightpath. There are three major contributors
to the energy consumption of this network architecture: (1) by optical
transponders (Erk ) (transponders at different line rates consume differ-
ent amounts of energy), (2) by optical amplifiers (Ea), and (3) by elec-
tronic processing (Ep). Optical switching of a wavelength channel in the
OXC consumes very little energy compared to other energy costs [142].
Therefore, in this work, we consider the optical switching energy cost to
be negligibly small.
5.2.2 Translucent Architecture
In a translucent IoWDM network, an optical signal can travel as far as
possible before the signal quality falls below a certain detectable thresh-
old [140]. So, the signal needs to be regenerated at an intermediate node
through O-E-O operation. Translucent networks employ signal regener-
109
Figure 5.2. Translucent IoWDM architecture.
ators at different nodes as needed during network planning [143], [144].
Between its source and destination, a signal can be regenerated several
times at intermediate nodes. The number of regenerators on a lightpath
depends on the signal reach and the length of the lightpath.
Figure 5.2 shows the architecture of a translucent IoWDM network.
Each node has an OXC connected wtih an IP router. Some nodes will also
be equipped with signal regenerators. As shown in Fig. 5.2, the reach
of a lightpath has been increased by placing regenerators at node B. In
addition to all the energy-consumption contributors to the transparent
IoWDM case, here the energy cost of regenerators (ERk) also contributes
to the energy consumption of the network.
5.2.3 Opaque Architecture
An opaque IoWDM architecture is on the other extreme of the fully trans-
parent IoWDM architecture. In an opaque network, both ends of a link
will have O-E-O interfaces at all nodes, thereby facilitating signal regen-
eration at every node [37], [144]. This basically allows signal regenera-
tion at every intermediate node in a lightpath. Essentially, a single hop
in an opaque IoWDM network is also of the same length of the physical
fiber link. Therefore, the opaque network architecture has a large num-
ber of O-E-O conversions, consequently increasing the network energy
consumption.
Figure 5.3 shows the architecture of an opaque IoWDM network. Here,
110
Figure 5.3. Opaque IoWDM architecture.
the IP router in connected to the OXC via Short-Reach Transponders
(SRT). The optical transponders in the OXC (referred to as Opaque Op-
tical Transponder (OOT)) do not perform O-E-O conversion, hence they
consume less energy compared to optical transponders in the transpar-
ent/translucent case. Therefore, along with optical amplifiers and elec-
tronic processing, the other two major contributors in the energy con-
sumption of opaque IoWDM networks are: energy consumption values of
SRT (ESRk) and OOT (Eork ). Number of OOTs in a lightpath can be easily
calculated using the number of intermediate nodes on the lightpath.
5.3 Energy-Efficient MLR Network ModelHere, we state the problem of designing an energy-efficient MLR IoWDM
network, a special version of which considering a single rate can model
a Single-Line-Rate (SLR) network. We present three different models for
designing energy-efficient transparent, translucent, and opaque IoWDM
networks. During variable indexing, we use the following rules: m and
n index the nodes in the physical topology of the network, i and j index
the nodes in the virtual lightpath topology, and s and d index source and
destination nodes of a traffic demand. The inputs to the design models
are as follows:
• G(V,E): A physical topology consisting of node set V and link set E.
At each node, an IP router is connected to an OXC. Node configu-
111
rations are further elaborated in the following subsections.
• T = [Λsd]: forecasted traffic matrix with aggregate demand Λsd be-
tween a (s, d) pair.
• R = r1, r2, ..., rk: set of available channel rates.
• W : maximum number of wavelengths supported on a link, λ ∈
{1, 2, ...,W}.
• Other inputs:
Erk: energy cost of a transponder with rate rk,
ERk: energy cost of a regenerator card with rate rk,
ESRk: energy cost of a SRT with rate rk,
Eork: energy cost of an OOT with rate rk,
Ea: energy cost of an amplifier, and
Ep: energy cost of electronic processing (per Gbps).
To describe the model, we introduce some more notations for the pa-
rameters and variables as follows:
• Lmn: length of fiber span between nodes m and n.
• Pmn: set of lightpaths through physical link (m,n).
• Amn: number of amplifiers on a fiber on link (m,n). If we are given
the span distance L (e.g., 80 km) between two neighboring ampli-
fiers (EDFA), the number of in-line amplifiers for a fiber link (m,n)
is given by Amn = ⌈Lmn/L− 1⌉+ 2 where 2 is used to count pre- and
post-amplifiers [37]. Obviously, the longer the link, the more the
number of amplifiers needed.
• Fmn: integer variable denoting the number of fibers on a physical
link (m,n).
112
• f sdij : integer variable denoting the volume of traffic from source s to
destination d on lightpath (i, j).
• Zj: integer variable expressing the amount of data carried by light-
paths which are terminated at node j.
Model-specific notations are provided in the respective subsections
below where we describe the models to design transparent, translucent,
and opaque MLR IoWDM networks.
5.3.1 Transparent IoWDM Network
We consider a transparent optical network with no wavelength conver-
sion and different reach for different line rates. In a node, the IP router
is connected to the OXC (e.g., ROADM) by long-haul transponders. More
notations to describe the model are as follows:
• lijkλ: lightpath between a node pair (i,j) at rate rk over wavelength λ.
• αijkλ: denotes whether a lightpath lijkλ is feasible or not based on an
acceptable BER threshold (B).
• Xijkλ: integer variable denoting the number of lightpaths (lijkλ) on
link (i, j) in the virtual topology at rate rk over wavelength λ.
The objective of the problem is to find the energy efficiency of a trans-
parent MLR network, and can be written as:
minimize 2∑λ
∑ij
∑k
Xijkλ.Erk +∑mn
Amn.Fmn.Ea +∑j
Zj.Ep (5.1)
subject to the following constraints:∑λ
∑k
rk.Xijkλ.αijkλ ≥∑sd
f sdij ∀(i, j) (5.2)
∑(i,j)∈Pmn
∑k
Xijkλ.αijkλ ≤ Fmn ∀(m,n),∀λ (5.3)
113
∑i f
sdij −
∑i f
sdji =
Λsd, if s = j
−Λsd, if d = j
0, otherwise
∀i, ∀(s, d)
(5.4)
Zj =∑sd
∑i
f sdij ∀i ̸= s (5.5)
The mathematical formulation of the problem turns out to be a mixed
integer linear program (MILP). The objective function (Eqn. (5.1)) min-
imizes the energy consumption of the MLR network. The first term in
Eqn. (5.1) computes the total energy consumption of WDM transponders
(2 for counting the source and destination transponders of a lightpath)
required to support the traffic demands. The second term calculates
the energy consumption of all the in-line amplifiers in the network. Fmn
quantifies the number of fibers needed to carry the traffic demands as we
may need multiple fibers on a link in case of low-bit-rate networks. How-
ever, we can easily downgrade the formulation for single-fiber networks
by forcing the value of Fmn to 1. The third term in Eqn. (5.1) captures
the total energy cost for electronic processing at each intermediate node
for all the traffic demands. We account here specifically for the traffic
that is electronically processed at intermediate nodes along multi-hop
lightpath routes, and we disregard the electronic processing of traffic at
source and destination nodes since this contribution is constant under
all the scenarios.
In the constraints, αijkλ determines whether the lightpaths Xijkλ be-
tween nodes i and j, at rate rk and on wavelength λ, are feasible based on
the BER threshold. These αijkλ values are calculated offline for each pos-
sible physical route. The physical routes are determined by any shortest-
path algorithm. In Eqns. (5.2) and (5.3), the multiplication of Xijkλ and
αijkλ ensures that only feasible lightpaths are present in the solution.
114
Equation (5.2) is the capacity constraint which limits the traffic de-
mands routed over a link (i, j) (in the virtual topology) by its capacity.
Equation (5.3) is the wavelength-continuity constraint which ensures
that, on a physical link with multiple fibers, there should not be more
than one lightpath on the same wavelength, i.e., there is no color clash.
Equation (5.4) is the flow conservation constraint which captures the fact
that, in all nodes of the network, total outgoing traffic should be equal
to total incoming traffic except for the source and destination nodes. If
an end-to-end traffic flow from i to j is routed using two lightpaths (i, k)
and (k, j), then at node k, electronic processing of that flow is required.
Equation (5.5) calculates the aggregated traffic flow at each node which
needs electronic processing. We can generate the formulation for a SLR
network by enforcing a single value for rk in the MILP model.
5.3.2 Translucent IoWDM Network
In the translucent case, regenerators can only be placed in the nodes. A
node can host regenerator cards of different line rates if needed. These
regenerators will increase the reach of different line rates. There is no
limit on regenerator sites, hence almost all paths can be feasible at var-
ious line rates. A path is only infeasible on a certain line rate if there
exists a link on that path which has higher length than the reach of
the line rate. There is no extra wavelength conversion available in the
network. At each node, the IP router is connected to the OXC through
long-haul transponders. We introduce some more notations to describe
the model as follows:
• EijRk
: energy cost of the regenerator cards with rate rk on lightpath
(i, j). These values are pre-calculated for each feasible lightpath
(i, j) at different line rates using the physical topology and ERk.
• Xijkλ: integer variable denoting the number of lightpaths on link
115
(i, j) in the virtual topology at rate rk over wavelength λ.
The objective of the problem is to find the energy efficiency of a translu-
cent MLR network, and can be written as:
minimize∑λ
∑ij
∑k
Xijkλ.(2Erk + EijRk) +
∑mn
Amn.Fmn.Ea +∑j
Zj.Ep (5.6)
subject to the following constraints:∑λ
∑k
rk.Xijkλ ≥∑sd
f sdij ∀(i, j) (5.7)
∑(i,j)∈Pmn
∑k
Xijkλ ≤ Fmn ∀(m,n),∀λ (5.8)
∑i f
sdij −
∑i f
sdji =
Λsd, if s = j
−Λsd, if d = j
0, otherwise
∀i, ∀(s, d)
(5.9)
Zj =∑sd
∑i
f sdij ∀i ̸= s (5.10)
This mathematical formulation is also a mixed integer linear program
(MILP). The objective function (Eqn. (5.6)) minimizes the energy con-
sumption of the MLR translucent network. The first term in Eqn. (5.6)
computes the total energy consumption of WDM transponders and re-
generator cards required in the translucent network to support the traffic
demands. The second and third terms are the same as in Eqn. (5.1).
Equation (5.7) is the capacity constraint which gives the highest amount
of traffic that can be routed over a feasible link (i, j) in the virtual topol-
ogy. Equation (5.8) is the wavelength-continuity constraint to remove
color clash from the network, i.e, on any fiber link, there should not be
more than one lightpath on the same wavelength. Equations (5.9) and
(5.10) are the same as Eqns. (5.4) and (5.5).
116
5.3.3 Opaque IoWDM Network
In the Opaque IoWDM network, due to O-E-O conversion at each node,
full wavelength conversion is ensured. Therefore, there is no need to
consider wavelength continuity along a lightpath. The OOTs do not have
O-E-O functionality, hence they consume less energy compared to the
transparent/translucent optical transponders at different line rates. Be-
low, we describe the energy-efficient opaque IoWDM model which has
some more notations as follows:
• Eijork
: energy cost of opaque transponders (OOT) over a lightpath
between a node pair (i, j) with rate rk. Eijork
= 2(N − 1)Eork where N
is the number of nodes in lightpath (i, j).
•
γijk =
0, if ∃(m,n) ∈ Πij : BERkmn ≤ B
1, otherwise
where Πij: set of physical links traversed by lightpath (i, j).
• Xijk: integer variable denoting the number of lightpaths on link (i, j)
in the virtual topology at rate rk.
The objective of the problem is to find the energy efficiency of a opaque
MLR network, and can be written as:
minimize∑λ
∑ij
∑k
Xijk.(Eijork
+2ESRk)+
∑mn
Amn.Fmn.Ea+∑j
Zj.Ep (5.11)
subject to the following constraints:∑k
rk.Xijk.γijk ≥∑sd
f sdij ∀(i, j) (5.12)
∑(i,j)∈Pmn
∑k
Xijk.γijk ≤ Fmn.W ∀(m,n) (5.13)
117
∑i f
sdij −
∑i f
sdji =
Λsd, if s = j
−Λsd, if d = j
0, otherwise
∀i, ∀(s, d)
(5.14)
Zj =∑sd
∑i
f sdij ∀i ̸= s (5.15)
This is also a mixed integer linear program (MILP) formulation where
the objective function (Eqn. (5.11)) minimizes the energy consumption
of the MLR opaque network. The first term in Eqn. (5.11) computes
the total energy consumption of SRTs and OOTs required to support the
traffic demands. The second and third terms are same as in Eqn. (5.1).
In the constraints, γijk determines whether the lightpaths Xijk be-
tween nodes i and j, at rate rk, are feasible based on the BER threshold
(B). If there exists a physical link (m,n), over which the lightpath (i, j) has
been routed, whose length is higher that the reach of the signal at the
given line rate, the lightpath (i, j) becomes infeasible. These γijk values
are pre-calculated for each possible physical route.
Equation (5.12) is the capacity constraint which limits the traffic car-
ried by a link (i, j) (in the virtual topology) by it capacity. Equation (5.13)
is the wavelength capacity constraint which states that, on a physical
link with multiple fibers, the number of lightpaths (Xijk) are limited by
the number of available wavelengths. Equations (5.14) and (5.15) are
the same as Eqns. (5.4) and (5.5).
5.4 Illustrative Numerical ExamplesHere, we present illustrative results obtained from the mathematical
models of the three different architectures for both MLR and SLR net-
works. As mentioned in Section 5.1, mixing line rates reduces the trans-
mission reach of optical signals. Therefore, as a first step for evaluating
118
our design models, we need to determine the maximum transmission
reach of different line rates when they are mixed with other line-rate
channels.
5.4.1 Reach Estimation
In an MLR network, dispersion management can reduce the signal degra-
dations due to non-linear effects to extend the transmission reach [138].
Legacy networks are dispersion-minimized for 10G signals, i.e., 10G sig-
nals can be received at the receiver without any pre/post dispersion com-
pensation which ensures maximum transmission reach of 10G signals.
However, 40G, having smaller pulse width in the time domain, has a
lower dispersion tolerance compared to 10G signals. So, the reach of
40G will be affected if carried over a 10G Dispersion-Minimized Fiber
(DMF) [137]. Same applies for 100G signals on a 10G DMF. Therefore, a
transmission system dispersion-minimized for 10G signals is not optimal
for 40G or 100G signals.
In a SLR network, the network can be dispersion minimized for the
corresponding line rate (i.e., 10G/40G/100G). However, in a MLR net-
work, we need to select the dispersion map for a certain line rate and
then determine the transmission reach of all line rate signals. First, we
calculate the transmission reach of different line rate signals in a SLR
network. To do so, we consider three different types of practical fiber
systems with optimized dispersion maps for 10G, 40G, and 100G. For a
10G DMF, we have a dispersion map with 5% chromatic dispersion un-
dercompensation per span; and for 40G, we have a map with 2% chro-
matic dispersion undercompesation per span. For 100G DMFs, we have
100% chromatic dispersion, and all impairments due to dispersion are
compensated at the coherent receiver for 100G [145].
For 10G, a non-return-to-zero-on-off-keying (NRZ-OOK) system is used;
for 40G, a 50% return-to-zero differential qudrature phase-shift keying
119
(RZ-DQPSK) is used; and for 100G, a 50% RZ dual-polarized DQPSK is
used with coherent receiver, as typically dominant modulation formats
for these rates in commercially-available devices [145]. Other typical
parameters are: dispersion parameter is D = 17 ps/nm-km, nonlinear
index is γ = 2.6e-20, effective area for the fiber is Aeff = 86.6µm2, and
launch power is 0 dBm [145].
We consider 80-km fiber spans terminating in a DCF and an erbium-
doped fiber amplifier (EDFA). We solve the Nonlinear Schrodinger Equa-
tion (NLSE) using split-step Fourier method to determine the maximum
optical reach of each wavelength on different types of fibers [145]. The
reach values were estimated for a threshold BER of 10−3. The pure trans-
mission reach of 10G, 40G, and 100G are 1800 km, 2200 km, and 7000
km, respectively. Higher transmission reach of 100G is due to the co-
herent reception and advanced modulation format that are expected to
be incorporated in commercially-available 100G transponders. 10G and
40G line rates are also expected to have higher transmission reaches us-
ing those technologies. To determine the reach of various line rates in a
MLR network, we consider that the network is dispersion minimized for
10G (as in legacy systems). In this scenario, the maximum transmission
reach of 10G, 40G, and 100G are 1750 km, 1800 km, and 900 km, re-
spectively. We use these reach data to pre-calculate feasible lightpaths
for our models.
5.4.2 Results
For solving the MILP models, we use ILOG CPLEX software on an In-
tel Core 2 Duo machine with 4 Gigabyte RAM. To keep the solution time
feasible, we set the relative objective gap to 5%. Hereafter, the words “en-
ergy cost” mean the energy consumption values of the network/element.
Our network topology is the Pan-European Cost239 network with 11
nodes and 26 links (Fig. 5.4). The actual Cost239 topology (as shown in
120
Figure 5.4. Cost239 topology (link lengths in km).
Fig. 5.4) does not have long enough paths for which regenerators need
to be deployed. In the actual Cost239 topology, both transparent and
translucent design models would yield the same results as no regenera-
tors would be needed in the networks. Therefore, to study the effect of
regenerators (translucent case) on the network energy cost, we multiply
each link length in the Cost239 topology by 1.35x. We use the traffic de-
mand matrix as given in Table 5.1. The base traffic demands sum up to 1
Tbps, and can be multiplied with different load factors, e.g., multiplying
all elements in Table 5.1 by 5 gives us 5 Tbps of traffic, denoted as 5T.
At most, 16 wavelengths can be multiplexed in each fiber, and there is
no limit on the number of fibers in each link. Each link is bidirectional,
i.e., one set of fibers exists per direction.
Table 5.2 shows the energy costs for WDM transponders, regenerator
cards, SRTs, OOTs, amplifiers, and electronic processing (at each inter-
mediate node), and they are normalized to the energy cost of a 10G WDM
transponder, which is 35W [146]. 40G transponder’s energy consumption
varies quite a bit (from 73W to 130W) depending on vendors [146], [147].
We consider an in-between value (∼90W). 100G transponder values are
yet to be publicly available in the literature. However, following our cor-
121
Table 5.1. Base traffic matrix.
Node 1 2 3 4 5 6 7 8 9 10 11
1 0 1 1 3 1 1 1 35 1 1 1
2 1 0 5 14 40 1 1 10 3 2 3
3 1 5 0 16 24 1 1 5 3 1 2
4 3 14 16 0 6 2 2 21 81 9 9
5 1 40 24 6 0 1 11 6 11 1 2
6 1 1 1 2 1 0 1 1 1 1 1
7 1 1 1 2 11 1 0 1 1 1 1
8 35 10 5 21 6 1 1 0 6 2 5
9 1 3 3 81 11 1 1 6 0 51 6
10 1 2 1 9 1 1 1 2 51 0 81
11 1 3 2 9 2 1 1 5 6 81 0
respondence with industry professionals, we assume the given value in
Table 5.2. Regenerator cards have O-E-O conversion, hence they con-
sume the power of less than two transponders [147], [148]. We can a
extrapolate 100G regenerator card’s energy cost from a 100G transpon-
der’s energy cost.
SRT’s energy consumption value is a little less than long-reach WDM
transponder’s [146]. We can safely assume that ∼20% less energy is
used for opaque optical transponders (OOT) compared to regular WDM
transponders since there is no O-E-O conversion [148]. Regular EDFA
amplifiers consume only ∼8W [146]. However, an in-line amplifier’s box
contains two-stage amplifiers, dispersion compensation module, etc. Sim-
ilarly, the pre- or post- amplifiers consume more energy than regular
EDFA amplifiers. Therefore, by consulting various sources (such as
[148]), we consider amplifier energy consumption as ∼35W.
122
Table 5.2. Energy consumption values of network components.
Component Energy Consumption (Normalized)
10G 40G 100G
Transponder 1x [146] 2.6x [146], [147] 5.7x
Regenerator 1.4x [148] 3.6x [147] 8x
SRT 0.7x [146] 1.9x [146] 4.1x
OOT 0.8x 2.1x [37] 4.6x
Amplifier 1x [148]
Electronic Processing 0.5x/Gbps [34]
Estimating the energy cost for electronic processing of traffic at an
intermediate node is trickier. In an intermediate node, depending on the
interconnection between OXC and IP router, this energy cost will vary. An
accurate model based on real data can predict the electronic processing
energy cost. Modelling electronic processing energy cost can be a future
research topic. Here, we consider the energy consumption of electronic
processing as given in [34].
We note that high-data-rate devices have “volume discount”, i.e., en-
ergy cost of capacity scales less than linearly as capacity increases. For
illustration purposes, we use energy cost values of Table 5.2, however,
our model is general and can take any energy cost values.
Figures 5.5, 5.6, and 5.7 show the total power consumption of SLR
(link rates are either 10G, 40G, or 100G) and MLR networks for transpar-
ent, translucent, and opaque architectures, respectively. Component-
wise breakdowns of these results are given in Tables 5.3, 5.4, and 5.5.
The results show that an MLR network consumes less energy compared
to the SLR networks. As mentioned in Section 5.2.3, opaque networks
require more energy than transparent/translucent networks.
Transparent IoWDM: MLR networks can save from 3% upto 83% of en-
123
1 5 10 200
2000
4000
6000
8000
10000
12000
Aggregate Traffic (Tbps)
To
tal E
ner
gy
Co
st (
No
rmal
ized
) 10G SLR40G SLR100G SLRMLR
Figure 5.5. Energy cost comparison of transparent Cost239 networks.
1 5 10 200
2000
4000
6000
8000
10000
12000
Aggregate Traffic (Tbps)
To
tal E
ner
gy
Co
st (
No
rmal
ized
) 10G SLR40G SLR100G SLRMLR
Figure 5.6. Energy cost comparison of translucent Cost239 networks.
1 5 10 200
2000
4000
6000
8000
10000
12000
Aggregate Traffic (Tbps)
To
tal E
ner
gy
Co
st (
No
rmal
ized
) 10G SLR40G SLR100G SLRMLR
Figure 5.7. Energy cost comparison of opaque Cost239 networks.
124
Table 5.3. Energy consumption of various components of transpar-ent Cost239 networks for different traffic loads (normalized to 10Gtransponder’s energy consumption).
Network Component 1T 5T 10T 20T
10G SLR
Transponder 240.0 1020.0 1984.0 3968.0
Amplifier 250.0 712.0 1270.0 2254.0
Elec. Proc. 118.0 102.5 20.0 20.0
40G SLR
Transponder 213.2 738.4 1362.4 2776.8
Amplifier 236.0 411.0 560.0 812.0
Elec. Proc. 160.0 205.0 250.0 40.0
100G SLR
Transponder 262.2 752.4 1379.4 2451.0
Amplifier 257.0 417.0 449.0 516.0
Elec. Proc. 195.5 312.5 350.0 530.0
MLR
Transponder 204.0 720.4 1341.6 2627.6
Amplifier 203.0 400.0 557.0 742.0
Elec. Proc. 191.0 130.0 45.0 30.0
ergy cost compared to SLR networks (Fig. 5.5). It can be seen that, at
different traffic loads, different SLR networks’ energy consumption tend
to be closer to that of MLR network’s - from 1T to 10T traffic, 40G SLR
networks’ costs are closer to the cost of MLR networks, while at 20T traf-
fic, 100G SLR network’s cost is closer. The 10G legacy SLR network will
increase the energy cost at rapidly-growing rate for higher traffic loads.
As expected, 10G SLR networks have highest transponder and ampli-
fier energy costs, i.e., more fibers and transponders need to be deployed
to support the traffic demands, and 100G SLR networks have the least
amplifier energy costs (Table 5.3). An interesting finding in that the elec-
tronic processing energy cost is the highest for 100G SLR networks. For
the 100G SLR networks, as the maximum transmission reach of 100G
transponder is very high (7000 km), and the 100G transponder energy
125
Table 5.4. Energy consumption of various components of translu-cent Cost239 networks for different traffic loads (normalized to 10Gtransponder’s energy consumption).
Network Component 1T 5T 10T 20T
10G SLR
Transponder 258.0 1040.0 2020.0 4008.0
Regenerator 0.0 2.8 8.4 22.4
Amplifier 270.0 768.0 1252.0 2304.0
Elec. Proc. 105.5 62.5 50.0 20.0
40G SLR
Transponder 213.2 748.8 1372.8 2776.8
Regenerator 0.0 0.0 7.2 7.2
Amplifier 248.0 408.0 560.0 812.0
Elec. Proc. 164.5 207.5 250.0 60.0
100G SLR
Transponder 296.4 798.0 1379.4 2451.0
Regenerator 0.0 0.0 0.0 0.0
Amplifier 249.0 413.0 440.0 516.0
Elec. Proc. 205.5 280.0 360.0 530.0
MLR
Transponder 206.4 742.4 1364.8 2644.0
Regenerator 1.4 7.0 15.4 36.0
Amplifier 271.0 442.0 548.0 754.0
Elec. Proc. 113.5 82.5 40.0 30.0
cost is also very high, the mathematical model tries to use existing light-
paths as much as possible instead of setting up a new lightpath for each
traffic demand. Therefore, at higher traffic loads, the electronic pro-
cessing energy cost at intermediate nodes increases in the 100G SLR
networks. For the other types of networks, the cost of electronic process-
ing can not dominate the cost of setting up new lightpaths. Therefore,
in those cases, the electronic processing cost remains low compared to
other energy costs.
Translucent IoWDM: In the translucent case, MLR networks reduce en-
126
Table 5.5. Energy consumption of various components of opaqueCost239 networks for different traffic loads (normalized to 10Gtransponder’s energy consumption).
Network Component 1T 5T 10T 20T
10G SLR
SRT 170.8 729.4 1408.4 2800.0
OOT 294.4 1249.6 2483.2 4972.8
Amplifier 309.0 740.0 1284.0 2336.0
Elec. Proc. 96.5 100.0 30.0 0.0
40G SLR
SRT 152.0 539.6 1007.0 1960.8
OOT 235.2 873.6 1667.4 3309.6
Amplifier 285.0 449.0 560.0 782.0
Elec. Proc. 156.0 227.5 265.0 240.0
100G SLR
SRT 180.4 516.6 934.8 1771.2
OOT 257.6 855.6 1536.4 2990.0
Amplifier 249.0 416.0 476.0 516.0
Elec. Proc. 275.5 387.5 475.0 610.0
MLR
SRT 143.6 511.6 953.6 1845.2
OOT 230.0 874.4 1650.8 3169.6
Amplifier 295.0 458.0 508.0 686.0
Elec. Proc. 131.0 110.0 70.0 110.0
ergy cost in the range of 1% to 83% in comparison with SLR networks (Fig.
5.6). From 1T to 5T traffic, 40G SLR networks’ costs are closer to the cost
of MLR networks, while from 10T to 20T traffic, 100G SLR networks’ costs
are closer. Again, 10G SLR networks incur higher energy costs at higher
traffic loads. As in the transparent case, amplifier and transponder en-
ergy costs are the largest for 10G SLR networks. 100G SLR networks
do not require any regenerators for the Cost239 topology as the reach
of 100G is very high (compared to Cost239 path lengths) whereas other
types of networks require more regenerator cards at higher loads. We
127
may expect to see the use of more regenerators for a network topology
where the path lengths are longer than those of the Cost239 topology.
Another interesting point to note is that, as more regenerators are being
used at higher loads, the electronic processing energy costs are getting
lower (Table 5.4) since the regenerators are increasing the reach of op-
tical signals. The electronic processing energy costs are the highest for
100G SLR networks. The reason remains the same as in the transparent
case.
Opaque IoWDM: Opaque MLR networks reduce energy consumption in
the range of 1% to 74% compared to opaque SLR networks (Fig. 5.7).
From 1T to 5T traffic, 40G SLR networks’ costs are closer to the cost of
MLR networks, while from 10T to 20T traffic, 100G SLR networks’ costs
are closer. As in previous cases, 10G SLR networks’ energy costs grow
at a higher rate compared to other types of networks as traffic load in-
creases. OOT energy costs are around 2 times of SRT energy costs. It will
grow even larger if we consider another topology where the path lengths
are longer than Cost239 topology. As before, amplifier and transponder
(SRT and OOT combined) energy costs are the highest for 10G SLR net-
works and electronic processing energy costs are the highest for 100G
SLR networks (Table 5.5).
Now, let us focus on the MLR networks of three different architectures.
Transponders are major consumers of energy in all the architectures -
WDM transponders in transparent/translucent architectures and SRT
and OOT in opaque architecture. Let us look at how the rates of these
devices are distributed over different traffic loads.
Figure 5.8 presents the distribution of transponders of different rates
at various traffic loads in the transparent MLR Cost239 network. It
shows that the rate at which 10G transponder’s usage increases in flatter
compared to 40G or 100G transponder’s usage rate. 40G transponders
128
1 5 10 200
200
400
600
800
Aggregate Traffic (Tbps)
Nu
mb
er o
f T
ran
spo
nd
ers
10G40G100G
Figure 5.8. Transponder distribution in transparent MLR Cost239 net-work.
1 5 10 200
200
400
600
800
Aggregate Traffic (Tbps)
Nu
mb
er o
f T
ran
spo
nd
ers
10G40G100G
Figure 5.9. Transponder distribution in translucent MLR Cost239 net-work.
129
1 5 10 200
2
4
6
8
10
12
Aggregate Traffic (Tbps)
Nu
mb
er o
f R
egen
erat
ors
10G40G100G
Figure 5.10. Regenerator distribution in translucent MLR Cost239 net-work.
1 5 10 200
100
200
300
400
500
Aggregate Traffic (Tbps)
Nu
mb
er o
f S
RT
s
10G40G100G
Figure 5.11. SRT distribution in opaque MLR Cost239 network.
130
1 5 10 200
200
400
600
800
Aggregate Traffic (Tbps)
Nu
mb
er o
f O
OT
s
10G40G100G
Figure 5.12. OOT distribution in opaque MLR Cost239 network.
are being used the most at higher traffic loads. 100G transponder num-
bers are growing steadily at higher traffic loads. We can expect that 100G
transponder numbers will exceed 40G transponder numbers at certain
higher traffic load (not shown in the figure). Transponders in translu-
cent architecture show the same trend as in the transparent case (Fig.
5.9). Both SRTs and OOTs in the opaque architecture also show similar
trends (Figs. 5.11, 5.12). The regenerator card distribution at different
line rates in the translucent case is given in Fig. 5.10. Note that, as data
rate increases, high-data-rate regenerator cards are used more.
CapEx vs. Energy Consumption: It is worth noting that, in SLR net-
works, there is a close relation between Capital Expenditure (CapEx)-
minimized and energy-minimized design [37]: it turns out that energy-
minimized design is also CapEx-minimized. In MLR networks, we can
consider two CapEx models: (a) CapEx of only transponders, and (b)
CapEx of transponders and deployed fibers. If the CapEx of transpon-
ders exhibit “volume discount" as given in the energy consumption val-
ues, energy-minimized MLR network will also be CapEx-minimized for
both the models.
131
5.5 ConclusionWe investigated the energy costs of MLR and SLR networks. We de-
veloped mathematical models to design transparent, translucent, and
opaque IoWDM networks. We applied these models on a case-study net-
work with realistic energy cost parameters. We found that an MLR net-
works can improve the energy efficiency of all the three optical network
architectures, and MLR networks consume less energy compared to SLR
networks.
132
Chapter 6
Conclusion
6.1 Summary of the Research ContributionsEnergy efficiency in telecom networks is gaining significant attention
among the telecom networks researchers. In this dissertation, we de-
veloped novel methods and techniques to build energy-efficient next-
generation telecom networks. The algorithms, architectures, design meth-
ods, and results presented in the dissertation will assist researchers and
telecom service providers in developing networks in an energy-efficient
manner. In this chapter, we summarize the important contributions and
findings in the dissertation.
Chapter 2 presented a comprehensive survey of energy-efficient pro-
tocols and architectures proposed and standardized in the literature for
telecom networks. We specifically emphasized on optical telecom net-
works. We also reviewed protocols and applications deployed over op-
tical network infrastructures, such as data centers and grid applica-
tions. This work should work as a comprehensive reference for future re-
searchers focusing on energy-efficiency aspects of the telecom networks.
In Chapter 3, we described the detailed procedures for building a pro-
totype for a hybrid Wireless-Optical Broadband Access Network (WOBAN),
a novel and attractive solution for future high-bandwidth and cost-effect-
133
ive access networks. Performances of quad-play (e.g., voice, video, data,
and wireless) applications over WOBAN prototype were also demonstrated.
We also elaborated on the detrimental effect of many wireless hops on
network performance. This prototype facilitates programmability, re-
source sharing, and slice-based experimentation and can be instrumen-
tal for experimental research on next-generation hybrid, cross-domain
access networks, including energy-efficiency research.
As mentioned before, WOBAN can be cost-effective solution for future
access networks in terms of CapEx. It can also significantly reduce net-
work Operational Expenditure (OpEx) by its energy-efficient features. In
Chapter 4, we demonstrated how we can provide energy-efficient broad-
band access using a WOBAN. Our model defined a reference guideline for
designing energy-aware WOBAN. We also analyzed the impact of these
energy-aware design decisions on WOBAN’s performance. We showed
that, using appropriate design parameters, the impact on WOBAN’s per-
formance can be minimized while also achieving energy efficiency. Our
design methodologies applied on WOBAN can also be generalized so that
they are also applicable to other access networks such as PON variants.
In Chapter 5, we presented mathematical models to design energy-
efficient Mixed-Line-Rate (MLR) networks where a single link can have
diverse co-propagating line rate signals (i.e., 10/40/100G). We provided
three different models which can be applied to design transparent, transl-
ucent, and opaque MLR networks. We investigated the energy costs of
MLR and Single-Line-Rate (SLR) networks (where a single link has ho-
mogeneous line rate signals) based on our models using realistic energy
cost values. Our results indicated that MLR networks can reduce the
energy costs of networks compared to SLR networks.
134
6.2 Future Research DirectionsEnergy efficiency in telecom optical networks is an area of growing in-
terest among the research community, and many future extensions of
the research topics presented in this dissertation as well as many other
topics can be investigated in future. As mentioned in Chapter 5, consid-
ering energy conservation among the most important design objectives
(along with cost and performance) represents a paradigm shift in the
network design, traffic engineering, and network engineering research.
Many of the existing techniques for optical telecom networks investigated
and developed over the years (i.e., protection, traffic grooming, dynamic
bandwidth assignment, etc.) should be re-thought under this new per-
spective. In this section, we briefly outline some future research direc-
tions for building green telecom networks.
6.2.1 Core Networks
In core networks, we envision three main future research issues, i.e.
energy-efficient network provisioning, energy-efficient network architec-
ture design, and time-aware energy-saving schemes.
6.2.1.1 Energy-Efficient Network Provisioning
Since the potential traffic growth in telecom networks will directly result
in energy consumption growth, energy-efficient routing, grooming, and
wavelength assignment may help to reduce energy consumption caused
by the increase of traffic. As a preliminary step, in order to measure the
energy consumption caused by the traffic load more accurately, detailed
energy consumption model of different network architectures need to be
investigated. The energy consumption models will help us to assess the
energy consumption of novel energy-aware routing and grooming meth-
ods. Energy-aware traffic grooming also leads to reduced network op-
eration cost [67]. Devising energy-aware traffic grooming techniques for
135
various network architectures can be an important research problem. In
future, traffic engineering problems should also be reinvestigated keep-
ing energy efficiency in mind.
6.2.1.2 Energy-Efficient Network Architecture Design
Energy-efficient network architecture design is being recently investi-
gated by many researchers. In Chapter 5, we developed models to de-
sign energy-efficient MLR optical networks. In [64], the author proposed
a new concept of telecom network that decreases carbon footprint by
intelligent placement of data centers and routers at locations where re-
newable energy sources are abundant. The concept of “follow the wind,
follow the sun” helps us to decrease the telecom network’s carbon foot-
print. Inspired by this concept, more innovative network architectures
involving renewable energy sources can be envisioned. Some of the re-
search topics in this area can be - i) dynamic all-optical networks to ef-
ficiently utilize the benefits of remote renewable energy sources, ii) new
grids and data storage architectures with distributed storage locations
in renewable energy area, etc [64].
6.2.1.3 Time-Aware Energy-Saving Schemes
Traffic load of the core network varies during different hours of the day
[57]. Huge amount of energy is being wasted as all the network equip-
ment are on even during low traffic load. Energy-saving schemes can be
applied to core networks to exploit the diurnal traffic load characteristics.
Some research efforts focused on turning off a network element when it
is not in use [53]. However, uncoordinated shutting down of network
elements may cause connection interruption and poor network perfor-
mance. In future, dynamic and coordinated energy-saving schemes and
control systems may be designed and developed to take advantage of
the traffic varying during different hours of the day. These schemes will
guarantee the network availability and reliability even during the energy-
136
saving mode. Besides, these schemes should also be time-zone aware,
so as to meet the traffic load variation in different time zones.
6.2.2 Metro Networks
In current metro networks, Ethernet and WDM rings are the two most
commonly-used network technologies. Recently standardized energy-
efficient Ethernet protocol - IEEE P802.3az [45] - can be adapted to the
Ethernet metro networks and can become an interesting research area
for deploying energy-efficient Ethernet metro networks. In WDM ring
networks, similar to the core network, traffic load management, energy-
efficient network architecture, and time-aware energy-saving schemes
are important research issues.
6.2.3 Access Networks
The energy-saving methods in access networks can be quite diverse due
to the existence of different access network architectures. Here, we cite
some of the interesting future research issues for designing green optical
access networks.
6.2.3.1 WOBAN and Related Architectures
In Chapter 4, we demonstrated how WOBAN can exhibit energy savings
without having any detrimental effects on network performance. How-
ever, aggressive energy-saving mechanisms may also affect network per-
formance. In future, it will be interesting to perform an elaborate study
of network performance while energy-saving mechanisms are in place.
Such studies can also feature experiments in real network environments
using the WOBAN prototype developed in Chapter 3.
WOBAN architecture can also be realized using cellular networks (i.e.,
LTE or WiMAX) as the front-end. This novel architecture, called Cel-
lular Optical Integrated Network (COIN), can be an attractive solution
for future high-bandwidth cellular networks. It can also exhibit energy-
137
efficient network operations. To enable energy savings, the COIN archi-
tecture should be modular, easily reconfigurable, and adaptive to traffic
load. COIN will feature decoupled Base Station (BS) architecture (where
the Digital Unit (DU) and Radio-Frequency Unit (RU) are separated and
connected by optical fibers) which enables DU modules to be shared by
several RUs, thereby increasing resource sharing.
Dynamic allocation of radio resources to cell sectors based on traffic
load will make the resource management at RU more energy and band-
width efficient. A centralized resource management algorithm for energy
efficiency and resource virtualization for both RU and DU can be incor-
porated at the BS Controller in COIN. Intelligent network management,
efficient bandwidth provisioning, and reconfiguration methods to virtu-
alize the network resources in COIN (where users exhibit various network
usage behaviors at different hours of the day) need to be developed for
energy savings. This dynamic resource management will also improve
robustness and availability of COIN.
6.2.3.2 Resource Optimization in PON
It is evident that user traffic profiles in access networks exhibit signifi-
cant variation. This may lead to the underutilization of some resources
(i.e., wavelengths or time slots) at different hours of the day. It is possi-
ble to utilize this variable behavior traffic profile to intelligently minimize
resource usage in the network. In this way, energy can be optimized in
future PON systems. For example, we can minimize the number of wave-
lengths needed by deliberately assigning users with complementary net-
work usage behaviors on the same wavelength, thus using fewer wave-
lengths, but keeping all of them at high utilization at most hours of the
day. Fewer wavelengths means fewer resources consumed, which trans-
lates to energy savings. Preliminary work on this topic can be found in
[92]. Efficient bandwidth allocation algorithm can also be designed that
138
uses least number of wavelengths to serve all user requests. This algo-
rithm can further improve its intelligence by learning the usage patterns
of its assigned users.
6.2.3.3 Energy Efficiency in Long-Reach PON
In future access networks, serving all the users with a single central
server may no longer be the best solution, especially for systems such
as Long-Reach PON (LR-PON) that have a coverage area (of over 100 km)
compared to only a few km in traditional access networks. We can intro-
duce multiple OLTs in the LR-PON system, in which ONUs are served by
the nearest OLT instead of the remote OLT. These OLTs can also serve as
“backups” for one another. This scheme also means we can spend less
energy for transmission and reduce the complexity of OLTs, making the
energy consumption even lower.
Future multicast applications on LR-PON will require several ROADMs
in a ring to share wavelengths [26]. The division of each wavelength’s
launch power among the ROADMs should be a function of the num-
ber of users (using the wavelength) under each ROADM and thus will
vary. By designing an intelligent power-control algorithm in ROADMs,
it is possible to control launch power in a very efficient way and save
energy.
139
References
[1] P. Leisching and M. Pickavet, “Energy footprint of ICT: Forecastsand network solutions,” OFC/NFOEC’09, Workshop on EnergyFootprint of ICT: Forecast and Network Solutions, San Diego, CA,March 2009.
[2] G. Kramer, Ethernet Passive Optical Networks, McGraw-Hill, USA,2005.
[3] J. Baliga, R. Ayre, W. V. Sorin, K. Hinton, and R. S. Tucker, “Energyconsumption in access networks,” OFC/NFOEC’08, San Diego, CA,Feb. 2008.
[4] S. Sarkar, P. Chowdhury, S. Dixit, and B. Mukherjee, Broad-band Access Networks: Technologies and Deployment, ch. HybridWireless-Optical Broadband Access Network (WOBAN). Springer,2009.
[5] Y. Zhang, P. Chowdhury, M. Tornatore, and B.Mukherjee, “Energyefficiency in telecom optical networks,” IEEE Communications Sur-veys and Tutorials, vol. 12, no. 4, pp. 441–458, Oct. 2010.
[6] P. Chowdhury, B. Mukherjee, S. Sarkar, G. Kramer, and S. Dixit,“Hybrid wireless-optical broadband access network (WOBAN):prototype development and research challenges,” IEEE Network,vol. 23, no. 3, pp. 41–48, May 2009.
[7] P. Chowdhury, M. Tornatore, S. Sarkar, and B. Mukherjee, “Build-ing a green wireless-optical broadband access network (WOBAN),”IEEE/OSA Journal of Lightwave Technology (JLT), Special Issue onVery High Throughput Wireless over Fiber Technologies and Appli-cations, vol. 28, no. 16, pp. 2219–2229, Aug. 2010.
[8] P. Chowdhury, M. Tornatore, S. Sarkar, and B. Mukherjee, “To-wards green broadband access networks,” IEEE GLOBECOM’09,Honolulu, HI, Dec. 2009.
[9] P. Chowdhury, M. Tornatore, and B. Mukherjee, “On the energy ef-ficiency of mixed-line-rate networks,” OFC/NFOEC’10, San Diego,CA, Mar. 2010.
[10] Energy Information Administration Brochures, “Greenhousegases, climate change, and energy,” Online: http://www.eia.doe.
gov/bookshelf/brochures/greenhouse/Chapter1.htm, May 2008.
140
[11] “Report of the IUCN renowned thinkers meeting, the future ofsustainability: Re-thinking environment and development in thetwenty-first century,” Online: http://cmsdata.iucn.org/downloads/
iucn_future_of_sustanability.pdf, Jan. 2006.
[12] W. Vereecken, L. Deboosere, D. Colle, B. Vermeulen, M. Pickavet,B. Dhoedt, and P. Demeester, “Energy efficiency in telecommu-nication networks,” European Conference on Networks and Opti-cal Communications & Optical Cabling and Infrastructure (NOC’08),July 2008.
[13] M. Pickavet, W. Vereecken, S. Demeyer, P. Audenaert, D. Colle, C.Develder, and P. Demeester, “Contribution and role of network ar-chitectures in the footprint reduction of ICT,” NOC/OC&I’09, Val-ladolid, Spain, June 2009.
[14] M. Pickavet, W. Vereecken, S. Demeyer, P. Audenaert, B. Ver-meulen, C. Develder, D. Colle, B. Dhoedt, and P. Demeester,“Worldwide energy needs for lCT: the rise of power-aware network-ing,” IEEE International Symposium on Advanced Networks andTelecommunication Systems (ANTS’08), Bombay, India, Dec. 2008.
[15] C. Lange and A. Gladisch, “Energy consumption of telecommu-nication networks - a network operator’s view,” OFC/NFOEC’09,Workshop on Energy Footprint of ICT: Forecast and Network Solu-tions, San Diego, CA, March 2009.
[16] C. Lange, D. Kosiankowski, C. Gerlach, F.-J. Westphal, and A.Gladisch, “Energy consumption of telecommunication networks,”European Conference on Optical Communication (ECOC’09), Vi-enna, Austria, Sept. 2009.
[17] B. Nordman, “ICT, networks, and energy: The energy perspective,”OFC/NFOEC’09, Workshop on Energy Footprint of ICT: Forecast andNetwork Solutions, San Diego, CA, March 2009.
[18] B. Mukherjee, Optical WDM Networks, Springer, 2006.
[19] K. Zhu and B. Mukherjee, “Traffic grooming in optical WDM meshnetworks,” IEEE Journal in Selected Areas in Communications,vol. 20, no. 1, pp. 122–133, Jan. 2002.
[20] S. Ramamurthy and B. Mukherjee, “Survivable WDM meshnetworks. part I-protection,” IEEE INFOCOM’99, New York, NY,pp. 744–751, Mar. 1999.
141
[21] R. S. Tucker, “The role of optics and electronics in high-capacityrouters,” IEEE/OSA Journal of Lightwave Technology (JLT), vol. 24,no. 12, pp. 4655–4673, Dec. 2006.
[22] R. S. Tucker, “Green optical communications - part I: Energy limi-tations in transport,” IEEE Journal on Selected Topics in QuantumElectronics, Special Issue on Green Photonics, vol. 17, no. 2, pp. 1–16, April 2011.
[23] E. Bonetto, D. Cuda, G. A. Castillo, and F. Neri, “The role of arrayedwaveguide gratings in energy-efficient optical switching architec-tures,” OFC/NFOEC’10, San Diego, CA, Mar. 2010.
[24] E. Bonetto, L. Chiaraviglio, D. Cuda, G. A. Castillo, and F. Neri,“Optical technologies can improve the energy efficiency of net-works,” ECOC’09, Vienna, Austria, Sep. 2009.
[25] R. S. Tucker, “Green optical communications - part II: Energy lim-itations in networks,” IEEE Journal on Selected Topics in QuantumElectronics, Special Issue on Green Photonics, vol. 17, no. 2, pp. 1–14, April 2011.
[26] H. Song, Long-Reach Passive Optical Networks. PhD Dissertation,University of California, Davis, 2009.
[27] M. Jimeno, K. Christensen, and B. Nordman, “A network connec-tion proxy to enable hosts to sleep and save energy,” IEEE Inter-national Performance Computing and Communications Conference,pp. 101–110, Dec. 2008.
[28] C. Develder, M. Pickavet, B. Dhoedt, and P. Demeester, “A power-saving strategy for grids,” Gridnets’08, 2008.
[29] D. Zuckerman, “Green communications - management included,”IEEE International Conference on Communications Workshops(GreenComm’09), Dresden, Germany, June 2009.
[30] N. Ghani, S. Dixit, and T. S. Wang, “On IP-over-WDM integration,”IEEE Communications Magazine, vol. 38, no. 3, pp. 72–84, Mar.2000.
[31] J. Manchester, J. Anderson, B. Doshi, and S. Dravida, “IP overSONET,” IEEE Communications Magazine, vol. 36, no. 5, pp. 136–142, May 1998.
[32] “Ciena white paper - Optical Transport System - CoreStreamAgility,” Online: http://www.ciena.com/files/Optical_Transport_
System_PB.pdf.
142
[33] P. Chaitanya and D. Datta, “Some schemes for controlling gaindynamics of EDFA chains in wavelength-routed optical networks,”IEEE International Symposium on Advanced Networks and Telecom-munication Systems (ANTS’09), New Delhi, India, pp. 1–3, Dec.2009.
[34] R. S. Tucker, “Energy footprint of the network,” OFC/NFOEC’09,Workshop on Energy Footprint of ICT: Forecast and Network Solu-tions, San Diego, CA, March 2009.
[35] “Alcatel-Lucent 1675 Lambda Unite MSS datasheet,” Online:http://www.alcatel-lucent.com/wps/DocumentStreamerServlet?
LMSG_CABINET=Docs_and_Resource_Ctr&LMSG_CONTENT_FILE=Data_
Sheets/22131_1675_LambUnMSS_R_9.1_ds.pdf.
[36] “MRV Optical Cross Connect datasheet,” Online: http://www.mrv.
com/datasheets/MCC/PDF300/MRV-MCC-OCC_A4_HI.pdf.
[37] G. Shen and R. S. Tucker, “Energy-minimized design for IP overWDM networks,” IEEE/OSA Journal of Optical Communicationsand Networking, vol. 1, no. 1, pp. 176–186, June 2009.
[38] J. Baliga, K. Hinton, and R. S. Tucker, “Energy consumption of theInternet,” COIN - ACOFT’07, Melbourne, Australia, June 2007.
[39] J. Baliga, R. Ayre, K. Hinton, W. V. Sorin, and R. S. Tucker, “Energyconsumption in optical IP networks,” IEEE/OSA Journal of Light-wave Technology (JLT), vol. 27, no. 13, pp. 2391–2403, July 2009.
[40] “Ciena 3600 datasheet,” Online: http://www.ciena.com/products/
products_cn3600_technical_info.htm.
[41] “Ciena Select OADM datasheet,” Online: http://www.ciena.com/
files/Select_OADM_DS.pdf.
[42] Metro Ethernet Forum, “Metro ethernet services - a technicaloverview,” Online: http://metroethernetforum.org/PDF_Documents/
metro_ethernet_services.pdf.
[43] P. P. Iannone, K. C. Reichmann, M. Birk, N. J. Frigo, R. M.Derosier, D. Barbier, C. Cassagnettes, T. Garret, A. Verlucco, S.Perrier, and J. Philipsen, “A 160-km transparent metro WDM ringnetwork featuring cascaded erbium-doped waveguide amplifiers,”OFC/NFOEC’01, 2001.
[44] “ITU symposia on ICTs and climate change,” Online: http://www.
itu.int/ITU-T/worksem/climatechange/index.html, 2009.
143
[45] “IEEE P802.3az energy efficient ethernet task force,” Online: http://www.ieee802.org/3/az/index.html.
[46] B. Dugerdil, “ETSI green agenda,” Online: http://www.itu.int/
dms_pub/itu-t/oth/06/0F/T060F0000090032PDFE.pdf, 2009.
[47] “TIA environment,” Online: http://tiaonline.org/environment/.
[48] “ATIS NIPP,” Online: http://www.atis.org/0050/index.asp.
[49] “NIPP-TEE,” Online: http://www.atis.org/0050/tee.asp.
[50] A. Alimian, B. Nordman, and D. Kharitonov, Network and TelecomEquipment Energy and Performance Assessment - Test Proceduresand Measurement Methodology. ECR Initiative, 2008.
[51] Verizon, “TEEER metric quantification,” Online: http://www.
verizonnebs.com/TPRs/VZ-TPR-9207.pdf, Jan. 2009.
[52] L. Chiaraviglio, M. Mellia, and F. Neri, “Energy-aware networks:Reducing power consumption by switching off network elements,”Gruppo nazionale Telecomunicazioni e Teoria dell’Informazione(GTTI’08), Rome, Italy, May 2008.
[53] L. Chiaraviglio, M. Mellia, and F. Neri, “Reducing power consump-tion in backbone networks,” IEEE ICC’09, Dresden, Germany, June2009.
[54] L. Chiaraviglio, M. Mellia, and F. Neri, “Energy-aware backbonenetworks: a case study,” IEEE International Workshop on GreenCommunications (GreenComm’09), Dresden, Germany, June 2009.
[55] B. Sanso and H. Mellah, “On reliability, performance and Internetpower consumption,” IEEE DRCN’09, Washington, DC, Oct. 2009.
[56] H. Mellah and B. Sanso, “Review of facts, data and proposals for agreener Internet,” BroadNets’09, Madrid, Spain, Sept. 2009.
[57] C. Lange and A. Gladisch, “Energy efficiency limits of load adaptivenetworks,” OFC/NFOEC’10, San Diego, CA, Mar. 2010.
[58] F. Idzikowski, S. Orlowski, C. Raack, H. Woesner, and A. Wolisz,“Saving energy in IP-over-WDM networks by switching off linecards in low-demand scenarios,” Conference on Optical Network De-sign and Modeling (ONDM’10), Kyoto, Japan, Feb. 2010.
[59] Y. Zhang, M. Tornatore, P. Chowdhury, and B. Mukherjee, “Time-aware energy conservation in IP-over-WDM networks,” Photonics inSwitching’10, Monterey, CA, July 2010.
144
[60] L. Ceuppens, “Planning for energy efficiency: Networking in num-bers,” OFC/NFOEC’09, Workshop on Energy Footprint of ICT: Fore-cast and Network Solutions, San Diego, CA, March 2009.
[61] J. Chabarek, J. Sommers, P. Barford, C. Estan, D. Tsiang, and S.Wright, “Power awareness in network design and routing,” IEEEINFOCOM’08, Phoenix, AZ, pp. 1130–1138, 2008.
[62] M. Baldi and Y. Ofek, “Time for a greener Internet,” IEEE Interna-tional Conference on Communications Workshops (GreenComm’09),pp. 1–6, June 2009.
[63] S. Figuerola, M. Lemay, V. Reijs, M. Savoie, and B. St. Arnaud,“Converged optical network infrastructures in support of futureinternet and grid services using IaaS to reduce GHG emissions,”IEEE/OSA Journal of Lightwave Technology (JLT), vol. 27, no. 12,pp. 1941 – 1946, June 2009.
[64] B. St. Arnaud, “CANARIE: Research networks to help reduce globalwarming,” OFC/NFOEC’09, Workshop on Energy Footprint of ICT:Forecast and Network Solutions, San Diego, CA, March 2009.
[65] E. Yetginer and G. N. Rouskas, “Power efficient traffic grooming inoptical WDM networks,” IEEE GLOBECOM’09, Honolulu, HI, Nov.2009.
[66] M. Xia, M. Tornatore, Y. Zhang, P. Chowdhury, C. Martel, and B.Mukherjee, “Greening the optical backbone networks: a traffic en-gineering approach,” IEEE ICC’10, Cape Town, South Africa, 2010.
[67] M. Xia, M. Tornatore, Y. Zhang, P. Chowdhury, C. Martel, and B.Mukherjee, “Green provisioning for optical WDM Networks,” IEEEJournal on Selected Topics in Quantum Electronics, Special Issue onGreen Photonics, vol. 17, no. 2, pp. 1–9, April 2011.
[68] B. Puype, W. Vereecken, D. Colle, M. Pickavet, and P. De-meester, “Power reduction techniques in multilayer traffic engi-neering,” International Conference on Transparent Optical Networks(ICTON’09), Azores, Protugal, June 2009.
[69] S. Huang, D. Seshadri, and R. Dutta, “Traffic grooming: a chang-ing role in green optical networks,” IEEE GLOBECOM’09, Honolulu,HI, pp. 1–6, 2009.
[70] Y. Wu, L. Chiaraviglio, M. Mellia, and F. Neri, “Power-aware rout-ing and wavelength assignment in optical networks,” ECOC’09, Vi-enna, Austria, Sep. 2009.
145
[71] M. Hasan, F. Farahmand, and J. Jue, “Energy-awareness in dy-namic traffic grooming,” OFC/NFOEC’10, San Diego, CA, 2010.
[72] M. Hasan, F. Farahmand, A. Patel, and J. Jue, “Traffic groomingin green optical networks,” IEEE International Conference on Com-munications (ICC’10), Cape Town, South Africa, May 2010.
[73] C.-S. Li, Y. Ofek, and M. Yung, “Time-driven priority flow controlfor real-time heterogeneous Internetworking,” IEEE INFOCOM’96,San Francisco, CA, Mar. 1996.
[74] D. Grieco, A. Pattavina, and Y. Ofek, “Fractional lambda switchingfor flexible bandwidth provisioning in WDM networks: principlesand performance,” Photonic Network Communications, vol. 9, no. 3,pp. 281–296, May 2005.
[75] A. Cianfrani, V. Eramo, M. Listanti, M. Marazza, and E. Vittorini,“An energy saving routing algorithm for a green OSPF protocol,”IEEE INFOCOM’10, Traffic Management and Monitoring Workshop,San Diego, CA, March 2010.
[76] H. D. Saunders, “The Khazzoom-Brookes postulate and neoclassi-cal growth,” The Energy Journal, vol. 13, no. 4, pp. 130–148, Oct.1992.
[77] C. Lange and A. Gladisch, “On the energy consumption of FTTHaccess networks,” OFC/NFOEC’09, San Diego, CA, March 2009.
[78] J. Baliga, R. Ayre, K. Hinton, and R. S. Tucker, “Architectures forenergy-efficient IPTV networks,” OFC/NFOEC’09, San Diego, CA,2009.
[79] F. J. Effenberger, “Opportunities for power savings in opti-cal access,” Online: http://www.itu.int/dms-pub/itu-t/oth/09/05/
T09050000010006PDFE.pdf, Feb. 2008.
[80] J. Mandin, “EPON powersaving via sleep mode,” IEEE P802.3av10GEPON Task Force Meeting, Online: http://www.ieee802.org/3/
av/public/2008-09/3av_0809_mandin_4.pdf, Sept. 2008.
[81] S.-W. Wong, L. Valcarenghi, S.-H. Yen, D. R. Campelo, S. Ya-mashita, and L. G. Kazovsky, “Sleep mode for energy saving PONs:Advantages and drawbacks,” IEEE GLOBECOM Workshops’09,Honolulu, HI, pp. 1–6, Dec. 2009.
[82] S.-W. Wong, S.-H. Yen, P. Afshar, S. Yamashita, and L. G. Kazovsky,“Demonstration of energy conserving TDM-PON with sleep mode
146
ONU using fast clock recovery circuit,” OFC/NFOEC’10, San Diego,CA, Mar. 2010.
[83] R. Kubo, J. Kani, Y. Fujimoto, N. Yoshimoto, and K. Kumozaki,“Sleep and adaptive link rate control for power saving in 10G-EPONsystems,” IEEE GLOBECOM’09, Honolulu, HI, Dec. 2009.
[84] H. Yuanling and M. Hajduczenia, “Adjustable timer value forpower saving,” IEEE P802.3av 10GEPON Task Force Meet-ing, Online: http://www.ieee802.org/3/av/public/2008-09/3av_
0809_hajduczenia_power_Saving.pdf, Sept. 2008.
[85] GPON Power Conservation, “ITU-T G-Series Recommendations -Supplement 45 (G.sup45),” May 2009.
[86] “PMC-Sierra GPON ONT SOC,” Online: http://www.pmc-sierra.
com/products/details/pas7401/.
[87] DSL Forum, “DSL dominates global broadband subscribergrowth,” Technical Report, Online: http://www.broadband-forum.
org/news/download/pressreleeases/YE06_Release.pdf, Mar. 2007.
[88] “AlcaLu paints its MSAN green,” Online: http://www.lightreading.
com/document.asp?doc_id=178722#msgs, July 2009.
[89] P. Tsiaflakis, Y. Yi, M. Chiang, and M. Moonen, “Green DSL:Energy-efficient DSM,” IEEE ICC’09, Dresden, Germany, June2009.
[90] J. M. Cioffi, S. Jagannathan, W. Lee, H. Zou, A. Chowdhery, W.Rhee, G. Ginis, and P. Silverman, “Greener copper with dynamicspectrum management,” AccessNets’08, Las Vegas, NV, Oct. 2008.
[91] L. Humphrey, “Opportunities and techniques for power saving inDSL,” BT, UK, Online: http://www.itu.int/dmspub/itu-t/oth/09/
05/T09050000010003PDFE.pdf, Feb. 2008.
[92] L. Shi, S. S. Lee, and B. Mukherjee, “Energy-efficient long-reach passive optical network: A dynamic wavelength allocationscheme,” Photonics in Switching’10, Monterey, CA, July 2010.
[93] I. Cerutti, L. Valcarenghi, and P. Castoldi, “Power saving architec-tures for unidirectional WDM rings,” OFC/NFOEC’09, San Diego,CA, March 2009.
[94] K. Le, R. Bianchini, M. Martonosi, and T. D. Nguyen, “Cost andenergy-aware load distribution across data centers,” HotPower’09,Big Sky, MT, Oct. 2009.
147
[95] B. St. Arnaud, “CAnet - news,” Online: http://emperor.canarie.
ca/pipermail/news/2009/000720.html, June 2009.
[96] D. A. B. Miller, “Rationale and challenges for optical interconnectsto electronic chips,” Proceedings of the IEEE, vol. 88, no. 6, pp. 728–749, June 2000.
[97] J. G. Koomey, “Estimating total power consumption by servers inthe U.S. and the world,” Technical Report, Lawrence Berkeley Na-tional Laboratory, Feb. 2007.
[98] K. Kant, “Power control of high speed network interconnects indata centers,” Workshop on High-Speed Networks (HSN’09), 2009.
[99] Green Computing, “Cisco, EMC team with MITto launch $100m green data center,” Online:http://www.greenercomputing.com/news/2009/06/11/
cisco-emc-team-mit-launch-100m-green-data-center, June 2009.
[100] “IBM, Syracuse University, New York State to build one of theworld’s most energy-efficient data centers,” Online: http://www-03.ibm.com/press/us/en/pressrelease/27612.wss, May 2009.
[101] P. Mahadevan, P. Sharma, S. Banerjee, and P. Ranganathan, “En-ergy aware network operations,” IEEE Global Internet Symposium(in conjunction with IEEE INFOCOM’09), Rio de Janeiro, Brazil, Apr.2009.
[102] P. Mahadevan, P. Sharma, S. Banerjee, and P. Ranganathan, “Apower benchmarking framework for network devices,” IFIP Net-working Conference, Aachen, Germany, May 2009.
[103] K. J. Christensen, C. Gunaratne, B. Nordman, and A. D. George,“The next frontier for communications networks: power man-agement,” Elsevier Computer Communications, vol. 27, no. 18,pp. 1758–1770, Aug. 2004.
[104] M. Gupta and S. Singh, “Greening of the Internet,” ACM SIG-COMM’03, Karlsruhe, Germany, pp. 19–26, Aug. 2003.
[105] W. Vereecken, L. Deboosere, P. Simoens, B. Vermeulen, D. Colle,C. Develder, M. Pickavet, B. Dhoedt, and P. Demeester, “Energyefficiency in thin client solutions,” International ICST Workshop onGreen Grids, Athens, Greece, Sep. 2009.
[106] L. Irish and K. Christensen, “A “green TCP/IP” to reduce elec-tricity consumed by computers,” IEEE Southeastcon, Orlando, FL,pp. 302–305, Apr. 1998.
148
[107] I. Foster and C. Kesselman, The grid: blueprint for a new computinginfrastructure, Morgan Kaufmann Publishers, 2004.
[108] National LambdaRail, “The network for advanced research and in-novation,” Online: http://www.nlr.net/.
[109] W. Feng, M. K. Gardner, G. Karmous-Edwards, J. Sobieski, andM. Veeraraghavan, “When optical networking meets grid comput-ing?” International Conference on Computer Communications andNetworks (ICCCN’06), Oct. 2006.
[110] A. Jukan and G. Karmous-Edwards, “Optical control plane forthe grid community,” IEEE Communications Surveys and Tutorials,vol. 9, no. 3, pp. 30–44, July 2007.
[111] S. Tomic and A. Jukan, “Performance analysis of infrastructureservice provision with GMPLS-based traffic engineering,” IEEEJournal on Selected Areas in Communications, vol. 25, no. 5,pp. 881–894, June 2007.
[112] Y. Li, Y. Yang, M. Ma, and L. Zhou, “A hybrid load balancing strat-egy of sequential tasks for grid computing environments,” FutureGeneration Computer Systems, vol. 25, no. 8, pp. 819–828, Sept.2009.
[113] A.-C. Orgerie, L. Lefevre, and J.-P. Gelas, “How an experimentalgrid is used: The Grid5000 case and its impact on energy usage,”IEEE Intnerational Symposium on Cluster Computing and the Grid(CCGrid’08), May 2008.
[114] S. Sarkar, S. Dixit, and B. Mukherjee, “Hybrid Wireless-OpticalBroadband Access Network (WOBAN): A review of relevant chal-lenges,” IEEE/OSA Journal of Lightwave Technology (JLT), vol. 25,no. 11, pp. 3329–3340, Nov. 2007.
[115] IEEE/OSA Journal of Lightwave Technology (JLT), Special Issue onConvergence of Optical Wireless Access Networks, Nov. 2007.
[116] J. Hu, D. Qian, H. Yang, T. Wang, S. Weinstein, M. Cvijetic, and S.Nakamura, “Triple play services over a converged optical/wirelessnetwork,” OFC’06, Anaheim, CA, Mar. 2006.
[117] W.-T. Shaw, S.-W. Wong, N. Cheng, K. Balasubramaian, X. Qing,M. Maier, and L. G. Kazovsky, “Hybrid architecture and integratedrouting in a scalable optical-wireless access network,” IEEE/OSAJournal of Lightwave Technology (JLT), vol. 25, no. 11, pp. 3443–3451, Nov. 2007.
149
[118] Z. Jia, J. Yu, and G.-K. Chang, “A full-duplex radio-over-fiber sys-tem based on optical carrier suppression and reuse,” IEEE Photon-ics Technology Letters, vol. 18, no. 16, pp. 1726–1728, Aug. 2006.
[119] G. Smith, A. Chaturvedi, A. Mishra, and S. Banerjee, “Wireless vir-tualization on commodity 802.11 hardware,” WINTECH’07, Mon-treal, Quebec, Canada, pp. 75–81, Sept. 2007.
[120] “ITU-T P.800.1 Mean Opinion Score (MOS),” Online: http://www.
itu.int/itudoc/itu-t/aap/sg12aap/history/p800.1/p800.1.html.
[121] G. Kramer, B. Mukherjee, and G. Pesavento, “Interleaved Pollingwith Adaptive Cycle Time (IPACT): A dynamic bandwidth distribu-tion scheme in an optical access network,” Photonic Network Com-munications, vol. 4, no. 1, pp. 89–107, Jan. 2002.
[122] P. Lin, T. Wang, Y. Suemura, and C. Qiao, “Improving accessperformance with an integrated PON and WiMAX with MIMO,”OFC/NFOEC’07, pp. 1–3, Mar. 2007.
[123] J. Russo, “Network technology energy efficiency,” Berkeley Sym-posium on Energy Efficient Electronic Systems, Berkeley, CA, June2009.
[124] C. Jones, K. Sivalingam, P. Agrawal, and J. Chen, “A survey ofenergy efficient network protocols for wireless networks,” WirelessNetworks, vol. 7, no. 4, no. 4, pp. 343–358, 2001.
[125] G. Kramer. Personal Communication, Mar. 2009.
[126] “NEC GE-PON data sheets,” Online: http://www.nec.co.jp/.
[127] V. Ramamurthi, A. S. Reaz, and B. Mukherjee, “Optimal capacityallocation in wireless mesh networks,” IEEE GLOBECOM’08, NewOrleans, Dec. 2008.
[128] M. Roughan and C. Kalmanek, “Pragmatic modeling of broadbandaccess traffic,” Computer Communications, vol. 26, no. 8, pp. 506–526, May 2003.
[129] R. Garcia, V. Garcia, X. G. Paneda, D. Melendi, and J. Perez, “Anal-ysis and modelling of a broadband fiber access network with highpeer-to-peer traffic load,” Simulation Modelling Practice and Theory,vol. 14, pp. 506–526, Oct. 2005.
[130] J. Farber, S. Bodamer, and J. Charzinski, “Measurement and mod-elling of Internet traffic at access networks,” EUNICE’98, Munich,pp. 196–203, Aug. 1998.
150
[131] “Netgear green,” Online: http://www.netgear.com.
[132] “ILOG CPLEX: High-performance software for mathematicalprogramming and optimization,” Online: http://www.ilog.com/
products/cplex/.
[133] “SMART 2020: Enabling the low carbon economy in the in-formation age,” GESI’08, Online: http://www.theclimategroup.
org/assets/resources/publications/Smart2020Report_lo_res.pdf,2008.
[134] “Facebook: Yes, We Need 100-GigE,” Light Reading, Online: http:
//www.lightreading.com/document.asp?doc_id=181899.
[135] A. Nag, M. Tornatore, and B. Mukherjee, “Optical network designwith mixed line rates and multiple modulation formats,” IEEE/OSAJournal of Lightwave Technology (JLT), vol. 28, no. 4, pp. 466–475,Feb. 2010.
[136] C. Meusburger, D. A. Schupke, and A. Lord, “Optimizing the migra-tion of channels with higher bitrates,” IEEE/OSA Journal of Light-wave Technology (JLT), vol. 27, no. 22, Nov. 2009.
[137] H. Bissessur, A. Hugbart, S. Ruggeri, and C. Bastide, “40Gover 10G infrastructure - dispersion management issues,”OFC/NFOEC’05, Anaheim, CA, March 2005.
[138] S. Chandrasekhar and X. Liu, “Impact of channel plan and dis-persion map on hybrid DWDM transmission of 42.7-Gb/s DQPSKand 10.7-Gb/s OOK on 50-GHz grid,” IEEE Photonic TechnologyLetters, vol. 19, no. 22, pp. 1801–1803, Nov. 2007.
[139] B. Ramamurthy, H. Feng, D. Datta, J. P. Heritage, and B. Mukher-jee, “Transparent vs. opaque vs. translucent wavelength-routedoptical networks,” OFC/NFOEC’99, San Diego, CA, Feb. 1999.
[140] A. Nag and M. Tornatore, “Transparent vs. translucent optical net-work design with mixed line rates,” OFC/NFOEC’09, San Diego,CA, March 2009.
[141] A. Saleh, “Transparent optical networking in backbone networks,”OFC/NFOEC’00, Baltimore, MD, 2000.
[142] F. Musumeci, F. Vismara, V. Grkovic, M. Tornatore, and A. Pat-tavina, “On the energy efficiency of optical transport with timedriven switching,” IEEE ICC’11, Kyoto, Japan, June 2011.
151
[143] A. Saleh, “Island of transparency- an emerging reality inmultiwave- length optical networking,” IEEE/LEOS Summer Topi-cal Meeting on Broadband Opt. Net. and Tech., p. 36, July 1998.
[144] G. Rizzelli, G. Maier, R. Longo, and A. Pattavina, “Comparison ofopaque and translucent WDM networks with different regenerator-placement strategies under static and dynamic traffic,” Grupponazionale Telecomunicazioni e Teoria dell’Informazione (GTTI’10),2010.
[145] E. Ip, “Nonlinear compensation using backpropagation forpolarization-multiplexed transmission,” IEEE/OSA Journal ofLightwave Technology (JLT), vol. 28, no. 6, pp. 939–951, March2010.
[146] F. Idzikowski, “Power consumption of network elements in IP overWDM networks,” Technical Report, Telecommunication NetworksGroup (TKN), TU Berlin, Germany, July 2009.
[147] “Transmode TM-Series data sheet,” Online: http://www.transmode.
com/.
[148] W. V. Heddeghem, M. De Groote, W. Vereecken, D. Colle, M. Pick-avet, and P. Demeester, “Energy-efficiency in telecommunicationsnetworks: Link-by-link versus end-to-end grooming,” Optical Net-work Design and Modeling (ONDM’10), Kyoto, Japan, Feb. 2010.