Case studies in energy consumption of Internet services:

Kerry HintonCEET

University of MelbourneAustralia

2

The future energy efficiency gaps

• Current data growth rate >> traditional energy efficiency improvement rate

• Technology is not keeping up with traffic growth– May suffer an “energy bottleneck”

0

2

4

6

8

10

12

14

16

2014 2015 2016 2017 2018 2019 2020

Nor

mal

ised

incr

ease

Year

Global mobile data 57% pa

Global IP data 23% paTrend mobile efficiency 20%

Trend core efficiency 15%

Efficiency gaps

Cisco, 2015GreenTouch, 2015

3

Centre for Energy-Efficient Telecommunications• Research centre located in the University of Melbourne

• Launched in March 2011

• Partnership between Alcatel-Lucent, the University of Melbourne and Victorian State Government

– $10 million for 2011 to 2015

– Additional funding of $2 million has extended CEET to 1st July 2016

• World’s first research centre focusing on energy-efficient telecommunication technologies

• Focus on collaboration between business and academia

• Major contributor to GreenTouch international consortium

4

Case study 1: Wireless access to the cloud• Cloud services widely promoted as greener than

on-site facilities:– Cloud Computing – The IT Solution for the 21st Century

• Carbon Disclosure Project Study 2011

– Salesforce.com & the Environment

• WSP Environment & Energy 2011

• Strong case for enterprise private cloud

• What about the public cloud?– Apple iCloud

– Google drive

– Microsoft sky drive

5

Consumer, interactive wireless cloud

- “free Google online storage, so you can keep ..- anything.”- “Your files .. can be reached from any smartphone, tablet, or computer. - “.. wherever you go, your files follow..”

- “It’s like an extra hard drive that’s available from any of the devices you use. ”- “whether you’re on your laptop on your new tablet, or on your phone .. you can get to your files in OneDrive.”

- “automatically and wirelessly store your content”- “automatically and wirelessly push it to all your devices”

Source: www.apple.com/au/pr/products/icloud/icloud.html

Source: www.google.com/drive/

OneDrive

Source: windows.microsoft.com/en-us/windows-8/getting-started-onedrive-tutorial

6

1.78

1.65

0.30

0.38

0.01

0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

4.5

Ave

rage

po

we

r (

GW

)

Wireless cloud 2015 averagepower consumption

Metro & Core

Data Centre

WiFi hotspot

Home WiFi

4G

90%

10%

Wireless cloud power

4G LTE67%

HOME WiFi25%

WiFiHOTSPOT8%

User technology

JO

UL

ES

PE

R B

IT

1

10

100

1000

10000

Home WiFi WiFi hotspot

4G Network Data Centre

Energy efficiencyEnergy efficiency

WiFihotspot

Home WiFi+ PON

4G

Metro&

Core

DataCentre

InternetUser

Public data centre

CEET, 2013

7

Case study 2: Interactive cloud servicesx105

Source: Vishwanath et al. 2013

(AJAX)

Interactive cloud services

8

• Local processing (not online) 0.2 W

An order of magnitude less power than on-line

• Decision to use the cloud depends on ...

Power consumption of end-user device (high vs. low)

Network to access the cloud (wired vs. wireless)

Complexity of the task (less vs. more computation)

Exchange of data with cloud

Cloud services User goes online User already online

3G 5.9 W 2.2 W

WiFi 3.1 W 2.3 W

Ethernet 2.5 W 2.2 W

9

• Stunning growth of Facebook traffic:

– 240+ billion photos

– 350+ million photos added per day

– 750+ million photos were uploaded over New Year’s Eve

– 7000+ Tera-Byte memory added per month

• Facebook reports its annual data center energy consumption

Case study 3: Photo sharing via cloud

Then Now

10

Facebook eco-system

• Hot & Warm photos are distributed by a Content Delivery Network• Cold Photos are distributed directly from data centres

Data Centre

CDN

CDNAccess

Edge

Data CentreData Centre

Edge

Access

Access

Distant

Local

Core

11

Traffic (measurement)

• Packet analyser software utility (Wireshark)

• Further compression on photos in user browsers before uploading to Facebook

• Exchanged Bytes for a

5MB Photo: – Laptop (Ethernet, WiFi)

• Upload = 500KB

• Download = 200KB

– Smartphone (4G, WiFi)

• Upload = 1.1 MB

• Download = 120K 0

0.2

0.4

0.6

0.8

1

0 1 2 3 4 5 6 7 8 9 10

Ob

serv

ed

tra

ffic

(M

Byt

e)

Original size of photos (MByte)

Download

Upload

12

• Direct measurement : Power-mate (resolution of 10 mW)

• Uploading and downloading same 5 Mbyte photo

Energy: user device

9

10

11

12

13

14

15

16

0 20 40 60

Po

we

r (W

att)

Time (sec)

Pidle

Start End

9

10

11

12

13

14

15

16

0 5 10 15 20

Po

we

r (W

att)

Time (Sec)

Pidle

Start End

5MB photo Laptop Mobile Phone

Ethernet WiFi 4G WiFi

Upload 106 J 114 J 40 J 23 J

Download 23 J 33 J 18 J 8 J

Upload Download

13

Cellular44%

WiFi Hotspot

24%

Ethernet16%

Home WiFi16%

Users’ traffic profile

• 350+ million photos upload every day

• Users have 140 friends on average.

• For a new uploaded photo

– Assume 90% of friends wants to look at the photo (126 friends)

Source: Cisco VNI, Global Mobile Data

Traffic Forecast Update, 2012–2017

Source: Cisco The zettabyte era,

2012-2017

Friend access technologies

14

12 GWh 292 GWh

• Facebook 2012 total data centre IT energy : 516 GWh(Source: Facebook)

• Total network energy consumption: 304 GWh

• Photo sharing network energy 60% of FB total data centre IT energy

– Wireless (4G/LTE) access network is main energy consumption

Sharing online network energy

15

Case study 4: Cloud storage for “Digital Universe”

• Digital Universe: all the data created by humankind

Source: EMC Digital Universe – Research & Analysis by IDC (7th annual study) – April 2014

http://www.emc.com/leadership/digital-universe/2014iview/executive-summary.htm

16

How much power to store all this data?

• Today 40% of digital universe is stored in data centres

• Seagate: By 2020 60% will be stored in data centresSource: Seagate 2014

• Expect move toward more energyefficient storage technologies

• Will this save energy?High Perf.

HDD/SSHD

SSD

Capacity HDD

Near Online HDD(MAID – Massive Array of Idle Disks)

Offline Removable Media(tape, optical disks)

Warehoused Removable Media(tape to 30 years, optical disks to 1,000 years)

Tier 0

Tier 1

Tier 2

Tier 3

Tier 4

Tier 5

– Solid State Drive

– Solid State Hybrid Drive

– Hard DiskDrive

Incre

asing e

ne

rgy efficie

ncy

17

Data storage in 2020• Consider three scenarios for 2020:

1. Same storage Tiers as 2013 (BAU)

2. Better use of lower power storage Tiers

3. Best likely use of storage Tiers

30%40%

50%

12%

20%

22%52%

34%22%

5% 5% 5%1% 1% 1%

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

3.BAU-60% 4.Improved 5.Best Likely

Pro

po

rtio

n o

f St

ora

ge

Tier 0 - SSD

Tier 1 - Hi PerfHDD/SSHD

Tier 2 - Cap. HDD

Tier 3 - Idle HDD

Tier 4 - Removable

1.BAU 2.Better 3.BestSource: Horman & Campbell 2014

18

2020 Digital Universe data storage power

• Without a significant transition to lower power storage data storage power will almost triple

• Need to re-think data retention policies

40.23

28.40

15.84 19.64

0

5

10

15

20

25

30

35

40

45

2013 2020

Pow

er (G

W)

Year

BAU

Better

Best

Source: Horman & Campbell 2014

19

Case study 5: Internet of Things (IoT)

• IoT forecasts– 212 Billion connectable things by 2020

– 4.6 Petabyte /month in Global M2M traffic by 2019

– Data increasing at 71% increase per year

• Mainly small IPv6 packets

• Increasingly sophisticated “things”IoT eco-system

Gray et al. 2015

Dominant power consumption

Cisco VNI, 2015

IDC, 2014

20

Power Consumption Model

• From “thing” to terminal unit (central office)

PGateway PCPE PRN PTU

RN TUIoT Gateway CPE RN TU

RN TU

P PP P P X X

N N

Gray et al. 2015

Case Study 5

21

Results• VDSL2 is least power-efficient while Shared Wi-Fi (with PON) is most

efficient.

• LTE is mostly efficient for bitrates below 75 kb/s but least efficient above250 kb/s

1

10

100

1 10 100 1000

Pow

er C

on

sum

pti

on

(W

atts

)

IoT Data Access Rate (kb/s)

PON

Single UserWi-Fi

VDSL2

Shared Wi-Fi

PtP

LTE(40% of BS traffic Ave.)

LTE(20% of BS traffic Ave.)

LTE(100% of BS traffic Ave.)

Gray et al. 2015

BS = LTE base station

22

Case study 6: Energy efficiency, CO2 and “rebound”

• Will the energy efficiency of e-services, of itself, provide for carbon footprint abatement?

• “Jevon’s paradox” (Rebound effect)– Improving efficiency will stimulate economic activity & increase

power consumption

Reduce energyconsumption

Improve energyefficiency

Intended cycle

Reduce energy costs

Stimulate demand

Increase energy consumption

Jevon’s

Paradox cycle

Suesβpeck, 2015

23

Australia’s National Broadband Network

• Nationwide broadband network

– Multiple technologies

– Covers entire nation

– Completion date: early 2020’s

• Replace hi-CO2 materialservices with lo-CO2

e-services

Access technology % premises

FTTP: Fibre to the premises 24%

FTTN: Fibre to the node 30%

FTTdp/B: Fibre to the distribution

point or building11%

HFC: Hybrid fibre-coaxial cable 28%

FTTN (remote footprint) 1%

Fixed Wireless 3%

Satellite 3%

Total premises 100%Source: NBN Co. Strategic Review, Exhibit 4-2, p. 97.

24

Economic model• Use Computable General Equilibrium (CGE) economic model

– TERM at Centre of Policy Studies: Victoria University, Aust.

• Include added national debt to pay-off NBN construction

• Compare economy relative to “No-NBN” future

– FTTH in city centres & business areas

– Primarily DSL with some HFC elsewhere

• Two data rate scenarios

Scenario Down stream Up stream

Modest data rate > 12Mbps entertainment, < 2.5Mbps rest

< 5Mbps

High data rate 2.5-10 Mbps Higher Ed. &Telehealth,10-25 Mbps rest

≤ 10 Mbps

25

Six Broadband Services1. Cloud Computing for business.

– Savings depend on the ICT intensity in economy sector.

2. Electronic Commerce for business & government.

– Significant productivity improvements.

3. Online Higher Education

– MOOCs’ with some face-to-face supervision.

4. Telehealth Practice

– Telehealth for aged care, teleconsulting & teleconsulting.

– Broadband-enabled locations in regional areas

5. Teleworking improving productivity & enabling greater labour-force participation.

6. Household entertainment Services

26

-2 -1 0 1 2 3 4

Export volume

Real government consumption

Real investment

Average real wage

Aggregate employment

Consumer price index

Nominal household consumption

Real household consumption

Nominal GDP

Real GDP

Percentage change (%)

Cloud Computing

Electronic Commerce

Online Higher Education

Telehealth Practice

Teleworking

Online Entertainment

Debt-servicing Repayments

Long-Term: High Bitrate Requirements

27

Long-run carbon footprint

• Jevon’s Paradox applies: Increased CO2 due to 2% increase in economic activity

• Need policies to adopt lo-CO2 practices

-0.5 0 0.5 1

Cloud Computing

Online Higher Education

Electronic Commerce

Teleworking

Telehealth Practices

Online Entertainment

Transport

Debt-servicing repayments

NBN incremental energy use

Percentage change compared with BAU (%)

NB

N im

pact

are

as

Scenario: High Service RequirementsTypical Long-Run Year post NBN Deployment

Carbon dioxide equivalent

28

National Broadband Networks

• The NBN will boost the Australian economy by about 2% with increase in national welfare

– Telehealth practice and Teleworking will be most beneficial

• It takes more than better Entertainment to make the NBN economically beneficial

– But a couple of valuable services using the NBN’s increased capabilities will be enough to make the NBN worthwhile

• Ubiquitous broadband alone will not move an economy to lower carbon footprint

– Need “greening with ICT” policies

• For example GeSI: SMARTER 2030

29

Case study 7: Minimising network power• Does minimising network power require global controller?

– Enforce optimal solution on all parties

• Real networks have many independent players

– Service providers, carriers, regulators, users

• Use Game Theory

– Multiple service providers (players) use a common network

– Seek Nash Equilibrium for the game

• Each service provider strategy is weighted balance of:

– Minimising their power consumption

– Balancing the network load

– Minimising their service delay

• How close is Nash Equilibrium to globally controlled power?

30

• Partial USNET topology

– 4 user nodes (Input)

– 4 server nodes (players)

– 14 network nodes

– 50 predefined routes between users and server nodes

• Input traffic over a diurnal cycle

SimulationUser node

server nodenetwork node

31

Results• Provided

– Players share enough network resources

– Objectives are relatively aligned• Weights for three factors approximately equal

• Power consumption of Nash Equilibrium is close to the global solution

– Don’t needglobal controller

32

Power consumption of the global Internet

Year

15% p.a. technology improvement

Po

we

r C

on

su

mp

tio

n (

W)

109

1011

1010

108

1012

2010

2015

2020

Access (PON)

Global electricity supply (3% p.a.)

TotalTotal (2010 Technology)

Tucker, OFC 2011

33

GreenTouch

• Bell Labs initiated global research consortium including

– Industry

– Government

– Academic organizations

• Fundamental research to pre-competitive innovations

• 48 member organizations with 350+ leading scientists

• Recognized by the World Economic Forum as an industry-led best practice toward sustainability

• Finalist for Edison “Collective Disruption” Award 2016

• Launched in May 2010

• Final report on June 2015 (www.greentouch.org)

• Silver medallist “Collective Disruption”: Edison Awards 2016

Deliver by 2015 architectures, specifications and solutions and demonstrate key technologies to increase network energy efficiency by a factor 1000 compared to

2010

34

GreenTouch

• Energy efficiency technology roadmap for 2020:– Requires all network equipment to be the latest generation

– Will require all equipment to be upgraded by 2020

– This is financially challenging over a 5 year period

• However GreenTouch has shown that the technologies are there to secure major improvements

Efficiency improvement

Traffic growth(2010 to 2020)

Net energy reduction

(2010 to 2020)

Mobile Access 10,000x 89x 99%

Fixed Access (consumer) 254x 8x 97%

Core network 316x 12x 96%

GreenTouch, 2015

35

The longer term future

• Most of the energy efficiency gains are “once-of”– Can not be used time and again to get continual gains each year

• Expect traffic to continue increasing

• Need a new paradigm to make ICT perpetually sustainable into the future

Energy bottleneck?

M. Weldon, 2015

Energy efficient solution

Business as Usual

Wireless access

“once-of” solutions

New paradigm

Thank you