Going Green Introduction to Network energy- efficiency research Inder Monga Chief Technologist Energy Sciences Network, Scientific Networking Division Lawrence Berkeley National Lab January 16 th , 2013
Going Green Introduction to Network energy-efficiency research
Inder Monga
Chief Technologist
Energy Sciences Network, Scientific Networking Division
Lawrence Berkeley National Lab
January 16th, 2013
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Talk Objective: Three Questions
1. Why is network energy efficiency important as a research topic?
2. What is Greentouch doing, and why is it relevant?
3. What can network operators, administrators do right now?
Thanks to Greentouch, especially Thierry Klein, for their approval and Sharing slides to be presented at this meeting
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
The New World of Devices and Data
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science 4
2010 2015 202010-2
10-1
100
101
102
103
Traf
fic (T
b/s)
Year
Wireless Data
Internet Video
Wireless Voice
P2P
Data from: RHK, McKinsey-JPMorgan, AT&T, MINTS, Arbor, ALU, and Bell Labs Analysis: Linear regression on log(traffic growth rate) versus log(time) with Bayesian learning to compute uncertainty
North America
Traffic doubling every 2 years
• 40% per year • 30x in 10 years • 1000x in 20 years
Exponential growth in total traffic continues
ESnet is growing at 10x every 4 years
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Data Growth continues to be Worldwide
MORE DATA MEANS MORE
POWER
78 Mtons of CO2 5 000 000 towers = 5 000 000 000 people
without broadband
Today Future
§ 17.5 GigaWatts § ~ 9 Hoover Dams § ~ 15 nuclear power
plants
§ ~ 15M car emissions a year
§ ~ 150,000 Paris to New York round-trip flights
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science 6
Slow-Down in Technology
Network energy efficiency
only increasing at 10-15% per year
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Growing Network Energy Gap!
+27% INCREASE
ENERGY
Energy consumption in communications service provider (CSP) networks is forecast to increase by 27% from 2012 to 2016
=5th INTERNET
If the internet was a country: energy consumption is higher than Russia and a little less than Japan
HIGHEST COUNTRY
2010 2015 202010-2
10-1
100
101
102
103
Traf
fic (T
b/s)
Year
Wireless Data
Total Backbone
Internet Video
Wireless Voice
P2P
Data from: RHK, McKinsey-JPMorgan, AT&T, MINTS, Arbor, ALU, and Bell Labs Analysis: Linear regression on log(traffic growth rate) versus log(time) with Bayesian learning to compute uncertainty
North America
2010 2015 202010-2
10-1
100
101
102
103
Traf
fic (T
b/s)
Year
Wireless Data
Total Backbone
Internet Video
Wireless Voice
P2P
2010 2015 202010-2
10-1
100
101
102
103
Traf
fic (T
b/s)
Year
Wireless Data
Total Backbone
Internet Video
Wireless Voice
P2P
Data from: RHK, McKinsey-JPMorgan, AT&T, MINTS, Arbor, ALU, and Bell Labs Analysis: Linear regression on log(traffic growth rate) versus log(time) with Bayesian learning to compute uncertainty
North America
2005 2010 2015 2020
10
20
30
40
50
Growth
Y ea r
Mobile Data
Internet Backbone
Mobile Efficiency
WirelineEfficiency
Growing Gap!
Traffic
2005 2010 2015 2020
10
20
30
40
50
Growth
Y ea r
Mobile Data
Internet Backbone
Mobile Efficiency
WirelineEfficiency
Growing Gap!
Traffic
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Talk Objective: Three Questions
1. Why is network energy efficiency important as a research topic?
2. What is Greentouch doing, and why is it relevant?
3. What can network operators, administrators do right now?
Thanks to Greentouch, especially Thierry Klein, for their approval and Sharing slides to be presented at this meeting
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
By 2015, our goal is to deliver the architecture, specifications and roadmap — and demonstrate key components and technologies —
needed to increase network energy efficiency by a factor of 1000 from current levels.
2010 2015 20201E-4
1E-3
0.01
0.1
1
10
100
Effic
ienc
y (M
b/s/W
)
Year
1000x Target
Total Network: BAU
2010 2015 20201E-4
1E-3
0.01
0.1
1
10
100
Effic
ienc
y (M
b/s/W
)
Year2010 2015 2020
1E-4
1E-3
0.01
0.1
1
10
100
Effic
ienc
y (M
b/s/W
)
Year
1000x Target
Total Network: BAU
GREENTOUCH MISSION (www.greentouch.org)
Global research consortium representing industry, government and academic organizations
Launched in May 2010
52 member organizations
300 individual participants from 19 countries
25+ projects across wireless, wireline, routing, networking and optical transmission
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Over 15 research programs and 25 research projects underway
• Wireless and mobile communications
• Wireline access
• Core networks and optical transmission
• Services, applications and trends
New approaches being taken: • Devices and low power electronics / photonics
• Architectures, algorithms and protocols
• “Power-follows-load” intelligent management
• Service and energy optimized networks
Two major public demonstrations in wireless and fiber-to-the-home technologies
Establish and define common reference architecture and roadmap with strategic research directions
© 2012 GreenTouch ConsortiumGreenTouch Introduction | 2012
GREENTOUCH Status
0.0
2.0
4.0
6.0
8.0
10.0
12.0
14.0
2010 2012 2013 2014 2015 2016 2017 2018 2019 2020 2022
W/s
ubsc
riber
Wireless LANOLT(/subscriber)HGW processorWireline LAN (Eth.)PON digitalOE PON
GPONXGPON
EE HW design
Long reach
Virtual HGW
BI PON
Low power electronics
Transparent CPE
Low power Optics
Sleepmode 2
Sleepmode
Short Term Long TermMedium Term0.0
2.0
4.0
6.0
8.0
10.0
12.0
14.0
2010 2012 2013 2014 2015 2016 2017 2018 2019 2020 2022
W/s
ubsc
riber
Wireless LANOLT(/subscriber)HGW processorWireline LAN (Eth.)PON digitalOE PON
GPONXGPON
EE HW design
Long reach
Virtual HGW
BI PON
Low power electronics
Transparent CPE
Low power Optics
Sleepmode 2
Sleepmode
Short Term Long TermMedium Term
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Buffer
InputQueuing
ReceiveFwd
Engine
FabricInterface
OutputFwd
Engine
OutputQueuing
L2 BufferingOptics Framer
Buffer Mem
L2 BufferingOptics Framer Switch
Fabric
Switch Fabric
Switch Fabric
Fabric Interface
L1+L2 L3
Switch
18 Chip-to-chip Interconnects
Buffer
InputQueuing
ReceiveFwd
Engine
FabricInterface
OutputFwd
Engine
OutputQueuing
L2 BufferingOptics Framer
Buffer Mem
L2 BufferingOptics Framer Switch
Fabric
Switch Fabric
Switch Fabric
Fabric Interface
L1+L2 L3
Switch
18 Chip-to-chip Interconnects
Router T1600 (640Gb/s)
0
1,000
2,000
3,000
4,000
5,000
6,000
7,000
0% 20% 40% 60% 80% 100%
Load
W
Router T1600 (640Gb/s)
[From Kharitonov 2009]
D.Kharitonov, “Time-Domain Approach to Energy Efficiency: High-Performance Network Element Design” 2009 IEEE GLOBECOM Workshops
http://www.caida.org/research/traffic-analysis/pkt_size_distribution/graphs.xml
IPv4 Cumulative
INTERCONNECTS
PACKET SIZE
ENERGY DOES NOT
FOLLOW LOAD
Router Limitations
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Network equipment hardware (routers and switches)
• Architecture and components • Functions, features and dimensioning • Low energy technologies (including electronics,
photonics, etc) • Power measurements
Network topologies and architectures • Tradeoff between optical and electronic data transport • Optimal joint IP-optical network design • Packet versus circuit-switched architectures • Energy efficient and simplified routing
Integration of application and transport layers � Cross-layer optimization for efficient content distribution
• Traffic engineering - Bandwidth allocation & traffic grooming - Elimination of over-provisioning - Efficient protection and restoration - Multicasting, elimination of junk and redundant
traffic ….
• Network management, operation and control � Quality of service support � Network-wide reconfiguration and control of
network elements (offline or online). Holistic, end to end approach
� Protocols and algorithms for managing and controlling network elements
� Control and data plane � Energy and traffic monitoring
Focused on components, technologies, systems, algorithms and protocols at the data link layer (L2), the network layer (L3) and the transport layer (L4) as well as interactions with lower and higher layers and research efficiencies that can be obtained from cross-layer optimizations and
joint designs
Core Routing and Switching Research Group
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
§ Athens Information Technology (AIT)
§ Bell Labs (Chair: Thierry Klein)
§ Broadcom
§ Chunghwa Telecom
§ Columbia University
§ Dublin City University
§ Electronics and Telecommunication Research Institute (ETRI)
§ Energy Sciences Network / Lawrence Berkeley Labs
§ Politecnico di Milano
§ Freescale Semiconductor
§ Fujitsu
§ Huawei Technologies
§ IBBT
§ IIT Delhi
§ INRIA
28 members organizations with 67 individual members
§ KAIST
§ Karlsruhe Institute of Technology
§ Nippon Telegraph and Telephone Corporation
§ Politecnico di Torino
§ Samsung Advanced Institute of Technology (SAIT)
§ Seoul National University
§ University of Manchester
§ University of Melbourne
§ University College London
§ University of Cambridge
§ University of Leeds (Co-Chair: Jaafar Elmirghani)
§ University of New South Wales
§ University of Toronto
Membership from GT members
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Timeframe: • Consistent with GreenTouch timeframe • Algorithms, architectures and technologies that can be demonstrated by 2015
• With evolutionary improvements through 2020 • Applied to 2020 traffic • Comparison with 2010 traffic and 2010 technologies
Assumptions: • Consider the traditional IP packet data network framework • Alternative paths and technologies are possible, but more speculative and are not
expected to fit in the timeframe: • Optical burst switching • Content centric networking • Adiabatic switching, quantum dot cellular automata, ….
Background and Assumptions
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Contributing Members
SCORPION: Silicon Photonic Interconnect and single chip linecard
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Contributing Members
OPERA: Optimal End to End Allocation
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Shufflenetwork
Shufflenetwork
4x4 switch element
4x4 switch element
Inputshufflenetwork
Outputshuffle
network
GatingSOAs
4x4 switch element
4x4 switch element
4x4 switch element
4x4 switch element4x4
switch element
4x4 switch element
4x4 switch element
4x4 switch element4x4
switch element
4x4 switch element
IP layer WDM layer
Contributing Members
STAR: Switching and Transmission
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Contributing Members
ZEBRA: Zero Buffer Router Architecture
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Contributing Members
SEASON: Service Energy Aware Sustainable Optical Networks
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Talk Objective: Three Questions
1. Why is network energy efficiency important as a research topic?
2. What is Greentouch doing, and why is it relevant?
3. What can network operators, administrators do?
Thanks to Greentouch, especially Thierry Klein, for their approval and Sharing slides to be presented at this meeting
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Green Research in networking is HARD!
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Network is a significant part of power consumption for Network Operators Footprint of communication networks
0
50
100
150
200
250
300
350
400
2007 2012
Wor
ldw
ide
elec
trici
ty
cons
umpt
ion
(TW
h/y)
customer premises equipment
office network equipment
telecom operator networks
21
219 TWh
CAGR: 10.1% 354 TWh
Paper submitted for the Optics Express ECOC 2012 Special Issue “Worldwide electricity consumption of communication networks”
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Network is a significant part of power consumption for Network Operators Joint footprint
0%
1%
2%
3%
4%
5%
6%
7%
0
100
200
300
400
500
600
700
800
900
1000
2007 2008 2009 2010 2011 2012
Perc
enta
ge o
f tot
al w
orld
wid
e
elec
trici
ty c
onsu
mpt
ion
Elec
tric
ity c
onsu
mpt
ion
in IC
T (T
Wh/
year
)
data centres computers comm. netw.
26
CAGR: 7%
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Mantra
“you can’t manage, what you don’t measure”
OR “You can observe a lot by just watching.”
Yogi Berra
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
Monitoring and Visualization of Energy consumption in networks (MAVEN)
Goal:
• Establish a baseline power profile for end-to-end networking
• Pushing for efficiency and optimization, both at device-level and network-wide
• Expose live power dataset to researchers
ESnet5: First nationwide R&E network instrumented for live power monitoring
• 100G optical and routing layer
• Wide ranging traffic load: elephant and mice flows
• Tentative plan for comprehensive power data by Jan 2013
Figure: Visualization of energy (alpha version, unreleased) consumed by ESnet’s ANI prototype network.
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
GreenSONAR (GLIF/OGF) Paola Grosso (UvA), Inder Monga (ESnet) and Cees de Laat (UvA)
1. Measure Power Data, real-time (next presentation by Jon Dugan)
2. Apply NM & PerfSONAR methods and architecture to Green & Energy information
• What is the right model to represent and share this data?
3. Share data multi-domain!
4. Let researchers worldwide, use the data to build better, more-efficient network devices
Lawrence Berkeley National Laboratory U.S. Department of Energy | Office of Science
§ ICT networks are growing rapidly § Scaling networks is becoming more difficult § Bringing focus to energy efficiency
§ ICT and research communities are organizing to address challenges
§ Dramatic, holistic change, but over long term evolution § Cooperative organizations such as GreenTouch guiding evolution
§ Several promising research directions and initial results have been obtained
§ We can be providers of “real” data to the researchers
Conclusions