Transcript No Slide Title - The Computer Laboratory

•
•
INTelligent Energy awaRe NETworks.
An EPSRC UK Funded Project in Green Networks
Professor Jon Crowcroft, Cambridge
And Professor Jaafar Elmirghani
University of Cambridge, &
University of Leeds, UK
[email protected]
[email protected]
Outline
•
Network energy consumption trends
•
•
•
•
•
•
•
•
•
Energy efficient routing in optical networks
•
•
•
•
•
•
Motivation
Access network power consumption trends
Power usage efficiency (PUE)
Photonic versus electronic switching
Terminal versus network power consumption
Wired and wireless network energy consumption comparison
Power management approaches
Access, metro, core power consumption
Energy per bit
Energy Efficient Routing (EER)
Performance Evaluation of EER
Sleep cycles
Conclusions
References
2
Outline

Network energy consumption trends
•
•
•
•
•
•
•
•
•
Energy efficient routing in optical networks
•
•
•
•
•
Motivation
Access network power consumption trends
Power usage efficiency (PUE)
Photonic versus electronic switching
Terminal versus network power consumption
Wired and wireless network energy consumption comparison
Power management approaches
Access, metro, core power consumption
Energy per bit
Energy Efficient Routing (EER)
Performance Evaluation of EER
Sleep cycles
Co-Optimising
•
Co-Lo Data Center&Sustainable Energy Production Location
3
Motivation
•
Ministry of Internal Affairs and Communications Japan report
concluded that ICT equipment (routers, servers, PCs and
network systems) consumed 4% of the total electricity generated
all over Japan in 2006, a figure of 45,000,000 MWh. Over the
past five years the figure has grown by more than 20% [1].
•
The goal is to reduce this figure and its CO2 impact. It also has
to be observed that ICT can help reduce the ecological impact
by reducing journeys and introducing more efficient business
processes.
•
Studies indicate that for ICT equipment, 50% of CO2 emission is
due to the production stage, 45% due to the usage stage and 5%
due to the recycling/disposal stage [2]. Therefore it pays to
reduce the number of elements in the network and to design
architectures and protocols with the per-element usage in mind.
4
Access network power consumption trends
•
The figure shows NTT DoCoMo’s energy consumption for communication
equipment and number of 3G base stations [1]
•
Energy consumption increase is proportional to the number of installed base
stations
5
Power usage efficiency (PUE)
•
PUE is defined as the ratio of “total energy consumption” to “IT
equipment energy consumption” [1].
•
In addition to IT equipment, lighting and air conditioning are the
main contributors to the total energy.
•
A typical value of PUE is 1.7 [1].
•
Therefore it is worthwhile reducing the equipment, designing the
usage carefully, but also examining high temperature / uncooled
components.
•
Advanced hardware design and removing the cooling element
reduces CO2 emission by 20% to 40% it is estimated [3].
6
Photonic versus electronic switching
•
Photonic switching has much lower energy consumption
compared to electronic switching.
•
It has been shown that the power needed per bit for switching is
100 to 1000 times higher in an electronic semiconductor switch
as compared to a photonic switch [4].
7
Terminal versus network power consumption
•
Typical current mobile terminal power consumption is 0.83Wh
per day (including battery charger and terminal) [1].
•
The corresponding network power consumption is 120Wh [1].
•
The ratio is 150:1 and therefore the network power consumption
is the main contributor to CO2 and effort has to be directed at the
network primarily.
•
Significant research effort has gone into extending the mobile
terminal battery life by optimising and reducing its power
utilisation from 32Wh per day in 1990 to 0.83Wh per day in
2008, a factor of 38 [1, 5].
•
In comparison the network power consumption has received little
attention to date.
8
Wired and wireless network energy consumption
comparison
•
•
•
•
A university campus network has been reported to use 6% of the total
campus power [6]. This amount corresponds approximately to the
production of 13 tons of nitrogen oxides (a precursor to ozone), 35 tons of
sulphur dioxide and 5100 tons of carbon dioxide as a result of burning
fossil fuels. If the associated heating and cooling are added, network
energy consumption will double [6].
As a result of monitoring on campus networks [6], it was observed that the
power consumption of multi port hubs and switches is almost independent
of the number of devices connected to the hub or switch. For example a
Cisco switch consumed 40 watts with no devices connected and 42 watts
with 16 devices connected. Therefore it pays to reduce the number of
elements in the network.
It was found on campus [6] that the wired network consumed between
200,000 and 500,000 kWh per year, while the wireless network consumed
only 30,000 kWh. A factor of 10 difference between wired and wireless
which indicates that the focus has to be on the wired network for energy
saving.
In the US, ICT accounts for 3% of the total country energy consumption
9
[7, 8].
Power management approaches
•
Internet power usage has continued to increase over the past
decade due to (i) more absolute number of devices (ii) higher
active power of devices and (iii) more active hours of usage per
day [9].
•
Can shut off CPU and instruction level (nano to micro seconds),
inter-packet or intra-flow CPU halt or shut-off (micro to
millisecond) and inter-flow, the entire computer/communication
can be turned off (seconds to hours) [9].
10
Access, metro, core power consumption (1)
•
Energy consumption is rapidly developing into an environmental and
political concern [11, 12]. It has been studied in transport, buildings etc,
but less in telecom.
•
There is also concern about constructing and maintaining large data
centres and switching centres [13].
•
Therefore the question has been raised whether the Internet growth will
be constrained by power rather than bandwidth [11].
•
In [11] the conclusion reached is that photonic switching alone will not
solve the Internet energy consumption problem (ie need to look at the
overall picture including switching, routing protocols etc).
•
In some studies the extra power needed for cooling is assumed to be
equal to the power used by the equipment [11, 14].
•
In the access (based on PON) typical power consumption estimates are
10W for optical network units (ONU) and 100W for optical line terminal
which resides in an edge node and connects to several ONUS [11].
•
A typical edge router in the metro, for example Cisco 12816, consumes
4.21 kW [11, 15].
•
A typical core router, such as Cisco CRS-1 multishelf system with 92 Tb/s
full duplex switching capacity consumes 1020 kW [11, 15].
11
Access, metro, core power consumption (2)
•
WDM systems connecting the edge nodes to the core node
consume 1.5 kW for every 64 wavelengths [11, 16].
•
Typically one multiwavelength amplifier is required per fibre,
consuming around 6W [11, 16].
•
The WDM terminal systems connecting core nodes consume
811 W for every 176 channels, while each intermediate line
amplifier consumes 622 W for every 176 channels [11, 17].
12
Some more on motivation & techniques
•
•
Current estimates indicate that power consumption accounts for
about half the cost of ownership of communication networks [18,
19].
Energy saving in networks is possible due to two main reasons
[18]:
•
Networks are provisioned at present for the worst case scenario
and many times over provisioned (3 to 5 times). Therefore
varying the number of active elements and sections of a network
according to demand can save power.
•
The power consumption of the network at present remains
substantial even when the network elements are idle. Therefore
provisioning just the right amount and introducing sleep
operations during idle times can help.
13
Summary
•
Consider energy used in manufacturing as well as operation,
therefore reduce the number of network components.
•
Consider PUE, therefore uncooled components and systems are
attractive.
•
Photonic switching instead of electronic routing whenever
possible.
•
Network power consumption higher than that of the terminal.
•
Wired part still consumes more power than the wireless part.
•
Reduce the “over provisioning” whenever possible.
•
Introduce sleep modes and sleep cycles.
•
Power consumption can account for up to half of the operating
costs in networks.
14
Outline
•
Network energy consumption trends
•
•
•
•
•
•
•
•

Energy efficient routing in optical networks
•
•
•
•
•
•
Motivation
Access network power consumption trends
Power usage efficiency (PUE)
Photonic versus electronic switching
Terminal versus network power consumption
Wired and wireless network energy consumption comparison
Power management approaches
Access, metro, core power consumption
Energy per bit
Energy Efficient Routing (EER)
Performance Evaluation of EER
Sleep cycles
Conclusions
References
15
Energy efficient routing in optical networks
•
The energy costs of the network will grow as the amount of data on
the network increases.
•
As the network expands in its capacity, energy consumption in the
core network is an important concern for the networking industry.
•
Some of the possible approaches that can reduce the energy
consumption include,
• put to sleep some of the wavelength routed nodes and,
• at a network level consider changing routes during low traffic
periods.
•
These two approaches decrease the QoS and connectivity. Energy
consumption can be reduced so long as the QoS performance
remains within SLA
16
Energy efficient routing in optical networks
•
Using intelligent optical control planes, lightpaths (or
wavelength channels) can have dynamic route selection
polices.
•
By using an efficient optical control management mechanism,
network nodes (WRN) can be set to ON or OFF states.
•
During the OFF cycle the nodes, adopt a sleep mode, cutting
down the traffic routed through them. Traffic originating at the
node or destined to the node is handled
•
The energy reduction achieved due to a sleep cycle is at the
cost of decrease in QoS.
17
Energy per bit and WRN architecture used in the study
Wavelength routed node (WRN) used in the network architecture
18
Energy per Bit
•
The two different types of
energy associated with the
optical networks,
• Energy associated with
the transmission of one
optical bit over fiber,
• Energy consumed by a
router
(WRN)
for
switching an optical
signal.
Fig: Wavelength routed node (WRN)
used in the network architecture
• The energy associated with the transmission of 1 bit can be
expressed as
19
Energy per Bit
•
The power consumed in an optical network path is given by,
•
If an optical bit traverses H hops, with each hop consisting of k optical
inline amplifiers, then the total energy consumed due to WRN and
EDFAs is,
Pd  Psignal  PWRN  PEDFA
(WRN )
( EDFA )
( H  1) Ebit
 kHEbit
•
( H  1)
The energy per bit across a fiber of length Ln between the nodes
n,n+1 is given by,
(WRN )
( EDFA )
Ebit (n)  Ebit
 an Ebit
•
The energy required to transmit an optical bit across H hops is given
by,
(t )
Ebit
s, d   Ebit (n)
H
20
Energy per bit
21
Calculation of energy needed
22
Calculation of energy needed
23
Parameters used
24
Energy Efficient Routing (EER) Algorithm
•
We propose an Anycasting routing technique to minimize
the energy consumption in the optical network.
•
Anycasting is defined as the communication paradigm, in
which the user has the ability to choose a probable
destination from a group of possible destinations unlike
deciding it a-priori as in unicast.
•
An Anycast request is denoted as a two-tuple (s,Ds),
where s is the source node initiating a session and Ds is
set of probable destinations.
•
A sleep cycle is defined as the time duration in which a
WRN cuts off the traffic routed through it and adopts an
OFF state.
25
Energy Efficient Routing (EER) Algorithm
•
Definition: We denote the network element vector for a
link i as,
i 
NEVi   
 i 
•
The overall NEV for a route R, consisting of links {i, i + 1,
. . ., j − 1, j} is given by,
j
 j

NEVR  k ,  k 
k i
 k i

•
Definition: We define the threshold parameters for a
service () as,

( )
 th , th 
T
26
Energy Efficient Routing (EER)
Algorithm
Fig: Burst header packet fields used in the
EER algorithm
27
Performance Evaluation
•
The National Science Foundation (NSF) network topology
and the Italian network are considered in our study.
Random sleep modes were considered.
•
We have considered the service threshold vector as ,
•
T
( )
 [5.7,10]
T
NSF network
Italian Mesh Network (IMnet)
28
EER and SPR algorithms in NSFNet
Comparison of energy dissipation in the NSF network for Shortest Path
Routing (SPR) and Energy Efficient Routing (EER), under varying traffic.
29
Energy consumed and energy saved for different
anycasting orders (k), NSFNet
Average power consumption for each
lightpath in various anycast scenarios in
NSFNet
Average energy saving obtained due to
anycasting in NSFNet
30
Blocking probability NSFNet, Energy consumption at
different anycasting orders (k), IMNet
Average blocking probability in various
anycast scenarios in NSFNet
Average power consumption for each
lightpath in various anycast scenarios in
IMNet
31
Energy saved and blocking probability IMNet
Average energy saving obtained due to
anycasting in IMNet
Average blocking probability in various
anycast scenarios in IMNet
32
Power saving obtained with sleep modes in 4/1 NSFnet and
6/1 IMnet
Power saving obtained with sleep modes
in 4/1 NSFnet and 6/1 IMnet
33
Summary
Summary
•
We have computed the energy required to transmit a bit
in an optical channel.
•
We have evaluated the energy consumption based on
per hop parameters and node architecture.
•
Using anycasting communication and efficient BHP
signaling, we have minimized the energy consumption
in optical burst switched networks.
•
The energy saving is obtained without significantly
scarifying the QoS.
34
Outline
•
Network energy consumption trends
•
•
•
•
•
•
•
•
•
Energy efficient routing in optical networks
•
•
•

•
•
Motivation
Access network power consumption trends
Power usage efficiency (PUE)
Photonic versus electronic switching
Terminal versus network power consumption
Wired and wireless network energy consumption comparison
Power management approaches
Access, metro, core power consumption
Energy per bit
Energy Efficient Routing (EER)
Performance Evaluation of EER
Sleep cycles
Conclusions
References
35
Related Work
•
•
Different solutions have been proposed to save energy in optical networks
In previous work [1] a static sleep cycles algorithm was proposed to reduce
the amount of energy consumed in optical networks.
• The network is divided into clusters of nodes. Clusters are set to switch between
•
the ON and OFF modes statically.
Putting some nodes in sleep state means that some traffic flows will have to
take longer routes, i.e. energy is saved at the expense of QoS.
_____________________________________________________________________________________
[1] B.G.Bathula, J. M. H. Elmirghani, "Green Networks: Energy Efficient Design for Optical Networks," Proceedings of Sixth
IEEE/IFIP International Conference on Wireless and Optical Communications Networks (WOCN 2009), Apr. 2009, pp. 1-5. 36
Intelligent sleep cycles for energy efficiency
•
In this work we propose an intelligent sleep cycles algorithm where
nodes switch between the ON and OFF modes dynamically according
to the traffic flows in the network.
•
The major consideration for any energy saving solution is the trade off
between the amount of energy saved and the level of performance
degradation.
•
Dynamic sleep cycles are expected to save more energy while keeping
the network performance within acceptable levels.
•
When nodes go to sleep, they can still transmit and receive traffic but
they cannot route traffic.
•
Nodes monitor the traffic flows passing through. A monitoring window
period is defined.
•
If within the monitoring window the overall blocking probability is less
than a certain threshold, some nodes are selected to go to sleep
according to the traffic flow and their location in the network topology.
37
Intelligent sleep cycles for energy efficiency
• A node which is the only neighbour for another node cannot go to
sleep.
• If the network blocking probability exceeds the acceptable (service)
blocking probability threshold, the most recent node to sleep wakes
up to improve the blocking probability.
• In this algorithm, the blocking probability threshold is setup (within
SLA) to achieve a trade-off between the amount of energy saved
and the network performance.
38
Intelligent sleep cycles for energy efficiency
START
Is Simulation Time==
Traffic Monitoring Window Start
Time?
NO
YES
NO
NO
Network Blocking Probability
<
Blocking Probability Threshold
YES
Are there any sleeping
network nodes?
YES
Select the M nodes with the lowest
passing by traffic which do not result
in disconnecting any nodes when they
go off
Select the most recent
sleeping node to be wake up
Sleep the full resources of the M
selected nodes
Update the list of sleeping
nodes
Simulation Time==
Simulation End Time
NO
YES
End
39
Anycasting Algorithm
• In this work we assume a Grid scenario where OBS is
implemented as the switching technique and anycasting as the
routing paradigm.
• The anycast algorithm proposed in [1] is implemented. It is based
on selecting the Grid resources that achieve the lowest number of
hops to reduce the total amount of energy consumed.
• A more “static” OCS / dynamic OCS optical network can also be
considered,
______________________________________________________________________________________________________
[1] De Leenheer et. al “Anycast Algorithms Supporting Optical Burst Switched Grid Networks”, International Conference on
Networking and Services, p 6 pp., 2006.
40
Simulation Scenario
•
•
Simulations were conducted on the high-speed
Italian network topology.
Bolzano
Milano
All nodes were considered to be connected to
local area or access networks.
Verona
Trieste
Torino
Venezia
•
Bolgano
Genova
Five nodes are selected to serve as Grid resource
centres.
Fireze
Pisa
•
Ancona
Perugia
The network is assumed to deploy 64 data
channels and 2 control channels.
Rome
•
The wavelength rate is assumed to be 10 Gb/s.
•
We assume that the nodes are equipped with full
wavelength conversion capability.
Pescara
Napoli
Bari
Cagliari
Potbanza
•
Deflection routing is used to reduce the burst loss.
•
We assume an average burst size of 1 MB.
Catanzaro
Palermo
Catania
41
Simulation Scenario
• Traffic is assumed to follow a Poisson distribution.
• Traffic generated from the nodes is assumed to be asymmetric:
60% of the traffic is destined to Grid resources and 40% of the
traffic is destined to ordinary nodes.
• The simulation results evaluate two parameters: the amount of
energy saved and the burst blocking probability.
• The blocking probability threshold is assumed to be 0.1 and the
traffic monitoring window is assumed to be 0.2 seconds.
• A range of values of blocking probability and monitoring window
size were considered.
42
Performance Evaluation
Effect of the Intelligent Sleep Cycles Algorithm
Network Saved Energy VS Normalized Load
• The intelligent sleep cycles algorithm
• However at a network load equal 1,
there was no energy saving because
network blocking probability at that
load is expected to be higher than the
blocking
probability
threshold
(therefore the number of nodes that
can go to sleep is limited).
Network Saved Energy(KWH) per 3 secs
has succeeded to save an amount of
energy of about 9 KWH during the
simulation run course for 0.1≤L≤0.9.
12
Anycasting without Energy Saving Technique
Anycasting with Energy Saving Technique
10
8
6
4
2
0
0.1
0.2
0.3
0.4
0.5
0.6
Normalized Load
0.7
0.8
0.9
1
Effect of Intelligent Sleep Cycles Algorithm on
Network Saved Energy
Performance Evaluation
Effect of the Intelligent Sleep Cycles Algorithm
• There is no difference in the blocking
• There is a slight difference at very
low load (L=0.1) because when
nodes go to sleep, they don’t affect
the performance seriously due to the
low congestion at that load.
• The impact of the intelligent sleep
cycles algorithm
0.2≤L≤0.9.
is
obvious
at
Anycasting without Energy Saving Technique
Anycasting with Energy Saving Technique
0.2
Blocking Probability
probability between the two curves
at very high load (L=1). Because the
blocking probability is higher than
the blocking probability threshold
therefore no node goes to sleep.
Blocking Probability VS Normalized Load
0.25
0.15
0.1
0.05
0
0.1
0.2
0.3
0.4
0.5
0.6
Normalized Load
0.7
0.8
0.9
1
Effect of Intelligent Sleep Cycles Algorithm on
Network Blocking Probability
Performance Evaluation
Effect of the Blocking Probability Threshold
Network Saved Energy VS Normalized Load
the amount of energy saved is equal for
all blocking probability thresholds.
12
• However at high load (0.4≤L≤0.9), the
energy saved reduces as the blocking
probability
threshold
decreases,
because under a very low blocking
probability threshold, it is unlikely to
send nodes to sleep.
Network Saved Energy(KWH) per 3 secs
• It is obvious that at lower loads (L≤0.3),
14
Anycasting without Energy Saving
Anycasting with Energy Saving (Blocking Threshold=0.1)
Anycasting with Energy Saving (Blocking Threshold=0.05)
Anycasting with Energy Saving (Blocking Threshold=0.03)
10
8
6
4
2
0
0.1
0.2
0.3
0.4
0.6
0.5
Normalized Load
0.7
0.8
0.9
1
Effect of Blocking Probability Threshold on the
Network Saved Energy
Performance Evaluation
Effect of the Traffic Monitoring Window Size
Network Saved Energy VS Normalized Load
• The effect of traffic monitoring window
size on the energy saved can be
clearly noticed for 0.2≤L≤0.9. When
traffic
monitoring
window
size
decrease, the amount of saved energy
increases.
• Small monitoring window sizes allow
the exploitation of brief durations when
the network blocking probability is low
allowing more energy saving.
• However, having a very small window
size requires a very efficient routing
algorithm for quick response to
network changes.
Network Saved Energy (KWH) per 3 seconds
15
No Energy Saving
Energy Saving(Traffic Monitoring Window Size=0.1 secs)
Energy Saving(Traffic Monitoring Window Size=0.2 secs)
Energy Saving(Traffic Monitoring Window Size=0.5 secs)
10
5
0
0.1
0.2
0.3
0.4
0.5
0.6
Normalized Load
0.7
0.8
0.9
1
Effect of Traffic Monitoring Window Size on the
Network Saved Energy
Summary
• We have proposed an intelligent sleep cycles algorithm for
energy saving in optical networks.
• Simulation results have shown that the intelligent sleep cycles
algorithm has succeeded to save a considerable amount of
energy with a limited performance degradation, specially at
lower network loads.
• Under a lower blocking probability threshold, better
performance was achieved but less energy savings
were gained.
• The energy savings per year, is around 100 GWH.
47
Co-Lo Sustainable Energy&Data Centers
•
•
•
UK tidal/wind -> 10%
But intermittent
Well connected by fiber
•
•
Cheaper to move data (and code) than current
•
•
•
•
•
•
Safe control with a lot of current/voltage flying around:)
Not simple linear with distance
Tradeoff is complex (latency is bad for users, but not so bad for
many cloud systems whole system migration (fast xen migration over 1-10Gbps) is
doable
Data migration or live mirroring (or delta/synch) needs looking at
Youtube cite CPU for re-coding video/audio as more energy
intensive than storage
Not just a convex optimisation problem
•
Can’t just do like dual (ECN/Kelly)
Migration Metrics
•
Need to capture cost in energy of data v. electricity
•
•
•
Akamai used spot price - not useful for us
Possibly can capture as a linear programming problem
Other work (personal containers) allows us to migrate
web service data
•
Backend (sqlservers etc) less obvious
References
1.
2.
3.
4.
5.
6.
Etoh, Minoru; Ohya, Tomoyuki; Nakayama, Yuji, “Energy Consumption
Issues on Mobile Network Systems,” Energy Consumption Issues on
Mobile Network Systems, July 28 2008-Aug. 1 2008 Page(s):365 – 368.
T. Origuchi, T. Maeda, M. Yuito, Y. Takeshita, T. Sawada, S. Nishi, and
M. Tabata. Eco-efficiency evaluation of 3G Services. In Proc. 7th Int.
Conf. on EcoBalance, Nov. 2006.
W. Wu. Green bts gives fresh breath. Technical Report 38,
Communicate, Huawei Technologies, Feb. 2008.
http://www.huawei.com/publications/PublicationIndex.do.
H. Hinton. Photonic switching fabrics. IEEE Communications Magazine,
28(4):71–89, April 1990.
J. Paradiso and T. Starner. Energy scavenging for mobile & wireless
electronics. IEEE Pervasive Computing, 4(1):18–27, 2005.
Matthews, H.S.; Hendrickson, C.T.; Hui Min Chong; Woon Sien Loh,
“Energy impacts of wired and wireless networks,” Electronics and the
Environment, 2002 IEEE International Symposium on, 6-9 May 2002
Page(s):44 – 48.
50
References
7.
8.
9.
10.
11.
12.
13.
K. Kawamoto et al, “Electricity used by office equippment and network
equipment in the US, “ Energy – The International Journal, vol. 27, No. 3,
pp. 255-269, 2002.
J. Koomey, “Rebuttal to testimony on ‘Koyoto and the Internet: The energy
implications of the Digital Economy,” Berkeley, CA; Lawerence Berkeley
National Laboratory. LBNL-46509, 2000,
http://enduse.lbl.gov/projects/InfoTech.html
Kenneth J. Christensena, Chamara Gunaratnea, Bruce Nordmanb and
Alan D. George, “The next frontier for communications networks: power
management,” Computer Communications 27 (2004) 1758–1770.
See
http://wil.cs.caltech.edu/mwiki/index.php?title=Internet_energy_efficiency
for a very useful review and set of papers
J. Baliga, R. Ayre, K. Hinton and R. S. Tucker, “Photonic Switching and the
Energy Bottleneck,” Photonics in Switching, 2007, 19-22 Aug. 2007 pp.125
– 126.
M. Gupta and S. Singh, “Greening of the Internet,” ACM SIGCOMM,
Karlsruhe, Germany, Aug. 2003.
A. Vukovic, “Data Centers: Network Power Density Challenges,” ASHREA
Journal, vol. 47, p. 55, Apr. 2005.
51
References
14.
15.
16.
17.
18.
19.
K. Dunlap, “Cooling Audit for Identifying Potential Cooling Problems in Data
Centers,” 2006. http://www.apcmedia.com /salestools/VAVR5UGVCN_R2_EN.pdf
Cisco Data Sheets. [Online]. http://www.cisco.com
Lucent Technologies Data Sheets. [Online]. http://www.lucent.com/eon
Fujitsu Data Sheets. [Online]. http://www.fujitsu.com
Sergiu Nedevschi, Lucian Popa, Gianluca Iannaccone, Sylvia Ratnasamy
and David Wetherall, “Reducing Network Energy Consumption via RateAdaptation and Sleeping,” Technical Report No. UCB/EECS-2007-128
http://www.eecs.berkeley.edu/Pubs/TechRpts/2007/EECS-2007-128.html,
October 29, 2007.
E. Miranda and L. McGarry. Power/thermal impact of networking computing.
In Cisco System Research Symposium, August, 2006.
52
Recent publications related to energy and networks
•
Bathula, B. and Elmirghani, J.M.H., “Energy Efficient Optical Burst Switched (OBS)
Networks,” Proc. of IEEE Global Telecommunications Conference (GLOBECOM’09),
Hawaii, 30 Nov. – 4 Dec, 2009.
•
Bathula, B. and Elmirghani, J.M.H., “Constraint Based Anycasting Over Optical Burst
Switched Networks”, IEEE OSA Journal of Optical Communication and Networking,
vol. 1, No. 2, pp. 35-43, 2009.
•
Bathula, B. and Elmirghani, J.M.H., “Green Networks: Energy Efficient Design for
Optical Networks,” Sixth IEEE International Conference on Wireless and Optical
Communications Networks (WOCN2009), April 2009.
•
Bathula, B. and Elmirghani, J.M.H., “Energy efficient architectures for optical
networks,” Proc IEEE London Communications Symposium, London, Sept. 2009.
•
Feng, W. and Elmirghani, J.M.H., “Lifetime evaluation in energy efficient rectangular
Ad-hoc wireless networks,” International Journal of Communication Systems, 2010.
•
Feng, W., Alshaer H. and Elmirghani, J.M.H., “Green ICT: Energy Efficiency in a
Motorway Model,” IET Communications, 2010.
•
Degala, S. and Elmirghani, J.M.H., “A Voronoi Based Energy Efficient Architecture for
Wireless Networks,” Proc. IEEE International Conference on Next Generation Mobile
Applications, Services and Technologies (NGMAST), 2009, pp. 377 – 382.
53