A Green InternetA Green Internet

Rod Tucker

ARC Special Research Centre for Ultra-Broadband Information Networks (CUBIN)

My Friends:Jayant Baliga, Kerry Hinton, Rob Ayre

Information Logistics and EnergyInformation Logistics and Energy

Metro/Edge Network

Core Network

Energy and the InternetEnergy and the Internet

Energy

In

Video Distribution

Network

Data Center

Access Network

Data

Out

• Greenhouse Impact• Energy-limited capacity bottlenecks (“hot spots”)

• OPEX

• Enabling energy efficiencies in other sectors

Why is Energy Important?

The Internet

Putting Things into ContextPutting Things into Context

“SMART 2020: Enabling the low carbon economy in the information age,”

GeSI, 2008 www.gesi.org

Information and Communication Technologies (ICT)

Global Emissions GtCo2

e

Global Emissions

Emissions

ICTOther

Business as Usual 2020

5% of total

GtCO2

e

2020 Abatements

SummarySummary

Estimating energy consumption of the Internet

Where is the energy consumed-

Core, metro, access network, data centres?

Information Logistics -

Cloud Computing

-

Travel Replacement-

Airmail vs. Internet

The Khazzoom-Brookes postulate

Caveat: “Making predictions is difficult –

especially about the future.”

EthernetSwitch

OLTSplitter

Cabinet

Metro/Edge NetworkCore Network

Network Energy ModelNetwork Energy Model

Edge Routers PON

Fiber

EDFA

Core Router

Optical cross connect

Video Distribution Network

Storage

ServerServer

Storage

Data Center

Fiber

Access Network

DSLDSLAM

Cu

OLT

SwitchONU

Cabinet

PtP

FTTN

DSLAM

Fiber

Cu

Broadband Network Gateways

Hot spots

Tier 1 Network

OversubscriptionOversubscription

Peak access rate sold to userM =

Average access rate

Oversubscription~ 3 Mb/s in 2009

~ 0.12 Mb/s in 2009

M = 25

2500

Pow

er (W

/use

r)

Baliga et al., 2008

0

20

25

5

Peak Access Rate (Mb/s)

15

100 20015050

10

Power Consumption of IP NetworkPower Consumption of IP Network

Total

Routers

Access (PON)SDH/WDM Links

Today’s Internet (~ 3 Mb/s)

2009 Technology

Oversubscription = 25

% o

f Ele

ctric

ity S

uppl

y

0.5

1.0

Traffic Growth = 40% p.a

2009 2020

Power Consumption in Access NetworksPower Consumption in Access Networks

Wave7 ONT-G1000i

Splitter GPON

PtP

Edge Node

Cisco 6513

Hitachi 1220

NEC AM3160

Cisco 4503

NEC VF200F6

TC Communications TC3300

Axxcelera

ExcelMax

Cabinet

Access N/W

FTTN VDSL2

Cu

Fiber

Fiber

Cisco uBR10012

RF Gateway Cisco DCP3000

HFCCu

RF Amp

Node

WiMAX

AxxceleraExcelMax BTS

Fiber

Hitachi 1220

Splitter

30

Peak Access Rate (Mb/s)

Pow

er P

er U

ser (

W)

1 10000

PtP

PON

100

20

10

10

32 Customers

M ~10

M = 1

M = 1

Power Consumption in Access NetworksPower Consumption in Access Networks

M = 1

PON FTTH is “greenest”

FTTN

WiMAXM= 10

20 users per sector

HFC

M= 10M= 1

M= Oversubscription

LowLow--Power States in User ModemsPower States in User Modems

“With implementation of this Code of Conduct,…

5.5 Millions tons of oil equivalent (TOE) will be saved per year.”

Off-State (W)

Low-Power State (W)

On-State (W)

ADSL-CPE 0.3 3.5 4.0

VDSL2-CPE 0.3 4.5 6.0

GPON ONU 0.3 5.0 9.0

PtP ONU 0.3 3.0 5.0

Extract:

EUROPEAN COMMISSIONDIRECTORATE -GENERAL JRCJOINT RESEARCH CENTREInstitute for the Environment and SustainabilityRenewable Energies Unit

Code of Conducton Energy Consumption of Broadband Equipment

Draft Version 3Issue 15 – 17 July 2008

Total

Access (PON)Routers

2.5 25 250 2500

Ener

gy p

er b

it (μ

J)

1.0

1000

0.01

10

0.1

100~1 μJ/b

WDM (Optical Fibre Trunk Links)

Peak Access Rate (Mb/s)

Network Network EnergyEnergy Consumption Consumption per Bitper Bit

~100 μJ/b

Ene

rgy

per b

it (n

J)

0.01

0.1

1

10

100

1000

OpticalAmp

PoSWDMTx/Rx

EthernetSwitch

CoreRouter

PONONU

(10 Mb/s)

Server

20092009

Energy per Bit in Network DevicesEnergy per Bit in Network Devices

Largest Routing System available today

Cisco CRSCisco CRS--1 Router1 Router

Buffers, 5%

Control Plane, 11%

Switch Fabric, 10% Power and Heat Management, 35%

I/O, 7%

IP Look-Up and Forwarding Engine,

32%

Linecard Chassis: 1.28 Tb/s, 13.6 kW

Switch Fabric Chassis: 8 kW Power consumption

Energy in Electronic Integrated CircuitsEnergy in Electronic Integrated Circuits

CMOS Gates

Cwire

⎡ ⎤= + ⎣ ⎦∑ ∑ 21Energy2gate wireE C V

=Power Energy x Bit Rate

CMOS ASICLinecard

Source: ITRS ’97-’06 Roadmaps

Switching Energy in CMOSSwitching Energy in CMOS

Moore’s Law: Number of transistors x 2 each 18 months

Sw

itchi

ng e

nerg

y E

s, J

10-20

10-18

10-16

10-14

2000 2010 2020

?

65

32

130

250

18

CMOS Feature size (nm)

40% decrease /18 months(40% efficiency gain p.a.)

Router Capacity GrowthRouter Capacity Growth

Based on G. Epps, CISCO, 2006

1

10

100

1000

10000

1994 1996 1998 2000 2002 2004 2006

X 19

93 P

erfo

rman

ce

2008

CRS-1 (~1.3 Tb/s,13.6 kW/rack)

12416 (~0.3 Tb/s/rack)

12016 (~80 Gb/s/rack)

Year

Moore’s law x 2/18 m

Router capacity x 2.5/18 m

CMOS energy efficiency 40% p.a.

2010

Router energy efficiency improving at 20% p.a. Neilson, JSTQE 2006

25000

Pow

er (W

/use

r)

% o

f Ele

ctric

ity S

uppl

y

Baliga et al., 2008

0

20

0.5

25

5

Peak Access Rate (Mb/s)

15

100 20015050

1.0

10

Effect of Efficiency GainsEffect of Efficiency Gains

Total

Routers

Access (PON)SDH/WDM Links

2020

Overall Technology Efficiency Improvement Rate

= 0% p.a

2500

Baliga et al., JLT, 2009

0

20

25

5

15

100 20015050

10

5% p.a

Overall Technology Efficiency Improvement Rate

= 0% p.a

20% p.aTarget

10% p.a

% o

f Ele

ctric

ity S

uppl

y

0.5

1.0

Effect of Efficiency GainsEffect of Efficiency Gains

2020Peak Access Rate (Mb/s)

Pow

er (W

/use

r)

•

Alternatives for complex PC

•

Sleep mode and fast “wake up”, adaptive modes of network and clients

•

Sleep mode

•

Adaptive control of components based on communication requirements

•

Adaptive sharing of content

•

Highly efficient power adaptive processing

•

Virtualization

Global Packet Core

Network

Global Packet Core

NetworkServers and Data Centres Application

Content ServerEfficient Clients

and home networks

Application

Content Server

Towards EnergyTowards Energy--Efficient NetworksEfficient Networks

After: Gladish

et al., ECOC 2008

•

Energy management solutions e.g. optical/electronic bypass

•

Content delivery networks (e.g. IPTV)

Access Network Access Network

Data CentersData Centers

1Koomey, 20082Revolutionizing Data Centre Efficiency—Key Analyses”McKinsey & Company, April 2008.

•

Data center electricity consumption is ~1% of the global total1

•

Energy consumption of data centers worldwide doubled between 2000 and 2006 2

•

Incremental US demand for data centre energy between 2008 and 2010 is the equivalent of 10 new power plants2

1 MW Data Center

“Burden factor”

of 1.8 to 2.5 for power consumption associated with cooling, conversion/distribution and lighting

Sources: EYP Mission Critical Facilities, Cisco IT, Network World, Customer Interviews, APC

0

10

20

30

40

50

Server Storage Network

50% 35% 15%

Energy Consumption in a Data CentreEnergy Consumption in a Data Centre

ICT Infrastructure Cooling

Conversion Loss

Lighting

Some typical issues/questions:

•

Wired vs. wireless broadband access

•

Can Cloud Computing save energy?

•

Renewable energy sources

•

Physical vs. Internet delivery of data

•

Can Internet Travel replacement save the planet?

•

Optimum design of Video on Demand systems

Information LogisticsInformation Logistics

Management of data, including, transport, storage, and processinManagement of data, including, transport, storage, and processingg

Core Network

Metro/Edge Network

OLTSplitter ONU

Fiber

EthernetSwitch

Edge Routers

Broadband Network Gateways

CoreRouter

OXC

Computer, Storage

Servers,Processors

StoragePublic Data Center

Cloud ComputingCloud Computing

Servers,Processors

StorageEnterprise Data Center

PON Access Network

Total Energy Consumption =

Energy/bit used in • Storage• Transport• Processing

XTotal number of bits

Cloud

•

Software-as-a-Service–

Stored on user’s computer with updates downloaded regularly

•

Service Bureau–

Most tasks done on lower end user machine, outsource the “big” jobs

•

Computing-as-a-Service–

Hosted and run on provider computer “farm” with data initially uploaded from user (Thin client model)

Can cloud computing save energy ?

Three Cloud Computing ScenariosThree Cloud Computing Scenarios

Metro/Edge Network

Service Center

Access Network

Computing Resource

Service Bureau – Hardware & Software

Here is my raw video footage

and edit commands

Mid-range Computers

Service BureauService Bureau

Metro/Edge Network

Service Center

Access Network

Computing Resource

Here is your DVD Image

Mid-range Computers

Service BureauService Bureau

Service Bureau – Hardware & Software

Energy Consumption of Service Bureau ModelEnergy Consumption of Service Bureau Model

Computer used for 20 hrs/week, plus some video encoding of ½ hour videos

Older Computer

Mid-range Computer

Low-end Computer & outsourcing

Transport energy dominates

0

20

40

60

80

100

120

140

160

180

0 20 40 60 80 100 120

Number of Encodings per Week

Ave

rage

Pow

er C

onsu

mpt

ion,

W

Following the Sun and the WindFollowing the Sun and the Wind

•

Locate computing and storage resources near sources of renewable

energy

•

Move data to follow the sun and winds

•

Requires many orders of magnitude increase in a data transport capacity

•

Transport/efficiency trade-off

Making best use of renewable energy

Data by Airmail vs. Data by the InternetData by Airmail vs. Data by the Internet

Data by Airmail:

Melbourne Sydney

3x105

32-GB flash

drives

The Internet

103

kg CO2

Data by Internet:

1000 Gb/s for 24 hours

105

kg CO2

Cargo Jet

(24 hours)

2 μJ/b

20 nJ/b

107

GB

107

GB

Total

Access (PON)Routers

2.5 25 250 2500

Ener

gy p

er b

it (μ

J)

1.0

1000

0.01

10

0.1

100~1 μJ/b

WDM Links

Peak Access Rate (Mb/s)

Energy Consumption Energy Consumption per Bitper Bit

~100 μJ/b

Flash chips by Airmail

2 μJ/b

Using the Internet for Travel ReplacementUsing the Internet for Travel Replacement

Source: CISCO, 2008

Video Conferencing

Travel Replacement Travel Replacement ––

Greenhouse Impact (C0Greenhouse Impact (C022 ))

~500 kg/person return

2 X 1 Gb/s for 6 hours = 4 TB

~5 kg/person

Air Travel

Video Conferencing

Melbourne Sydney

Business Meeting

0.1

1

10

100

1,000

10,000

100,000

100 1 10 100 1000

Mb/s-hr

Dis

tanc

e tra

velle

d (k

m)

CarTrain

Plane

Tele-work

Travel

10

Gb/s-hr

Bitrate-Time Product

Bicycle

Daily work

Business Meeting in Sydney

RodRod’’s Telecommute Calculators Telecommute Calculator

The The KhazzoomKhazzoom--Brookes PostulateBrookes Postulate

D. Khazzoom

Energy Journal,10,1987, L. Brookes Energy Policy, 20,1992 Inhaber, “Why

Energy Conservation Fails”, Quorum, 2002H. Herring, http://technology.open.ac.uk/eeru/staff/horace/kbpotl.htm

•

Energy efficiency at micro level reduction of energy use at this level

•

But leads to an increase in energy use, at the macro level

•

Example:

Wide bodied passenger aircraft lower costs per passenger large increase in air travel increased greenhouse emissions

“Energy efficiency is absolutely the wrong approach to network design.

The objective should be to make the network carbon neutral”

-

Bill St. Arnaud, CANARIE

Some Interest Groups and OrganizationsSome Interest Groups and Organizations

http://www.gesi.org/

http://www.thegreengrid.org/home

http://www.itu.int/climate

http://wattwatt.com/

http://ictandclimatechange.com/

http://www.atis.org/0050/

•

Energy consumption of the Internet is small, but growing

•

Internet energy consumption dominated by–

Access network today

–

Core network in the future

•

A multi-faceted approach needed to build a green Internet:–

Improved efficiency in electronic and photonic devices

–

Low-energy switching techniques–

Improved architectures

–

New protocols

•

Information Logistics–

A new approach to network design and management

Summary Summary –– The Way ForwardThe Way Forward

Melbourne Information Logistics Centre (MILC)?

Centre for Information LOgistics (CILO)?