Energy Efficient Wireless Internet Access

IMT Lucca, 25/11/2009 MAM

Energy Efficient Wireless Internet Access

Marco Ajmone Marsan, Michela Meo

Politecnico di Torino


WIA & MtCO2e

Marco Ajmone Marsan, Michela Meo

Politecnico di Torino


What’s all this “green networking” about?


• Energy is becoming the issue of our futureWe depend on energy which is becoming scarce

Energy consumption is causing dramatic climate changes

• We must cope with this and reduce energy consumption in all sectors,

ICT and networking included

The problem


Climate change

Source: Hansen, J., et al. (2006) "Global temperature change". Proc. Natl. Acad. Sci. 103: 14288-14293.


2003 Model2009 Model

Climate change

Source: A.P. Sokolov et al, “Probabilistic Forecast for 21st Century Climate Based on Uncertainties in Emissions (without Policy) and Climate Parameters”, Report 169, Jan 2009


• The main global warming culprit is carbon dioxide, CO2

• Gases that react to form smog

• Fine particles such as black carbon

• 80% of the increase of CO2 in the air in the last century is due to fossil fuel burning (20% deforestation)

Who is the culprit?


Source: Energy Information Administration (EIA), International Energy – Annual Energy Outlook 2009

TW


• Electricity = 30% of energy

• 1 W of electrical energy ≈ 2.1 W of primary energy

Source: Energy Information Administration (EIA), International Energy – Annual Energy Outlook 2009


• Information and Communication Technologies play a positive role for energy saving:– moving bits instead of atoms

• teleworking and telecommuting• e-commerce• intelligent transport systems• electronic billing

– sensors to monitor and manage environment

What about ICT?


The ICT sector is a heavy consumer!

… but

“ICT alone is responsible of a percentage which vary from 2% to 10% of the world power consumption.”

“Electricity demand of ICT is almost 11% of the overall final electricity consumption in Germany.”

“The ICT sector produces some 2 to 3% of total emissions of greenhouse gases.”


Source: M. Pickavet et al, “Worldwide Energy Needs for ICT: the Rise of Power-Aware Networking,” in IEEE ANTS Conference, Bombay, India, Dec. 2008.

Which ICT?


Consumption might double in the next decade

Source: M. Pickavet et al, “Worldwide Energy Needs for ICT: the Rise of Power-Aware Networking,” in IEEE ANTS Conference, Bombay, India, Dec. 2008.


• Life Cycle Assessment (LCA) refers to the quantitative characterization of the environmental impacts of products and services and includes– Manufacture– Operation – Disposal

• A life cycle perspective can lead to a better understanding of environmental management

This is particularly true for IT products

Life cycle matters


• Example: 2-gram memory chip requires – at least 1,200 grams of fossil fuels – 72 grams of chemicals

• Fossil fuels for production are some 600 times the weight of the chip (the total fossil fuel to produce a car is 1-2 times its weight)

• Purification to semiconductor grade materials is energy intensive

• Due to its extremely low-entropy, organized structure, the materials intensity of a microchip is orders of magnitude higher than that of “traditional” goods.

Electronics


-2000

0

2000

4000

6000

8000

10000

12000

14000

16000

Desktop Laptop CRT LCD

To

tal

En

erg

y (M

J)

Production

Distribution

Use

End of life

Source: Peter James and Lisa Hopkinson, “Energy and Environmental Impacts of Personal Computing -- A Best Practice Review prepared for the Joint Information Services Committee (JISC)”, May 2009.

PCs

Williams, E., 2004. Energy Intensity of Computer Manufacturing: Hybrid Assessment Combining Process and Economic Input-Output Methods. Environ. Sci. Technol., 2004, 38, 6166-6174.

Lawrence Berkeley National Laboratory, 2005. Optimization of Product Life Cycles to reduce Greenhouse Gases in California. Report for California Energy Commission. CEC-500-2005-110-F.

IVF Industrial Research and Development Corporation, 2007. Lot 3: Personal Computers (desktops and laptops) and Computer Monitors. Final Report for the European Commission, August 2007.


Equipment Consumption

Desktop PC 100-150W

Laptop PC 20W

Server 700 W – 10KW

Router 5-10 W per Gbps

GSM BS 700W

UMTS BS 800W

WIMAX BS 400W

Operation


Data centers

Source: “Report to Congress on Server and Data Center Energy Efficiency” Public Law 109-431. U.S. Environmental Protection Agency ENERGY STAR Program , August 2007


• In 2006, U.S. data centers used 61 TWh of electricity, corresponding to 1.5% of national consumption

• Double the amount consumed in 2000

• Based on current trends, energy consumption will continue to grow 12% per year, due to increasing demand for the services they provide

Data centers


• Algorithms to free up servers and put them into sleep mode or to manage load on the servers in a more energy-efficient way

• Sensors identify which servers would be best to shut down, based on environmental conditions

• Use more efficient components• Reduce cooling needs (cooling consumes as

much as 40% of the operating costs) through specific physical layouts

Current solutions


Data centers

Source: “Fact Sheet on National Data Center Energy Efficiency Information Program“, U.S. Department of Energy (DOE) and U.S. Environmental Protection Agency (EPA), March 19, 2008


Internet

Core

Backbone

Metro

FeederNetworks



Typical network

Source: J. Baliga, K. Hinton and R. Tucker, “Energy consumption of the Internet”, in COIN - ACOFT 2007, June 2007, Melbourme, Australia

factor 4factor 4


Source: R. Tucker et al., “Energy consumption in IP networks”, in European Conference on Optical Communication ECOC’2008, Brussels, Sept. 2008.

Routers



Fixed operators

70% of power consumption70% of power consumption



Mobile operators



Order of the OPEX!

michela

Qui il 90% e' il consumo della rete vs il 10% dei terminali, perche' i terminali sono gia' fatti per consumare molto poco (Di questo 90% poi l'80% e' all'accesso). Questo per far vedere la differenza con le reti fisse in cui il 70% e' all'utente ed essendo distribuito sugli utenti e' piu' difficile da andare a ridurlo in maniera efficace


Which business model?


Fast Slow

Intermediate


Cellular networks• Base stations are responsible for about 80% of

energy consumed by a cellular network• A typical BS consumes from 500W to 3KW, with an

average consumption per year of 35 MWh (as much as 10 families)

• In Italy 60,000 BSs, leading to 2.1 TWh/year, about 0.7 % of total Italian consumption of electricity

• 300 M€ electricity bill for the operators

• About 1,2 Mton of emitted CO2 equivalent per year


Base station consumption


An immediate solution for mobile operators

Start by reducing consumption at the

access network with current technologies


Dynamic network planning

• Networks are planned based on the peak hour traffic

• Due to natural traffic variability (i.e., typical day/night traffic profile) the network results over-dimensioned during long periods of time

Switch off portions of the network

when traffic is low


Traffic profiles


• Assume that a fraction x of the base stations (cells) are switched off

• The BSs that remain on are in charge of ― the traffic of the cells that are off (the

desired QoS must still be guaranteed)― the radio coverage (transmission power

might be increased to guarantee coverage)

Switch-off scheme


A NodeB controls 2 microcells


Switch off Switch off halfhalf of the NodeB, x=1/2 of the NodeB, x=1/2


Switch off Switch off halfhalf of the NodeB, x=1/2 of the NodeB, x=1/2


8:00 16:00 24:00 8:00 16:00 24:00 8:000

0.01

0.02

0.03

0.04

0.05

0.06

0.07

time

lam

bda

day/night traffic patternfor one cell

Low traffic threshold: QoS is guaranteed

Total traffic in x+1 cells

night zone


8:00 16:00 24:00 8:00 16:00 24:00 8:000

0.01

0.02

0.03

0.04

0.05

0.06

0.07

time

lam

bda

night zone

traffic pattern for cells

remaining on


3.Check the maximum cell radius, RMAX

If R’< RMAX DONE

else • increase transmission power during night

zone OR• reduce the night zone

Looking for a switch-off scheme


8:00 16:00 24:00 8:00 16:00 24:00 8:000

0.01

0.02

0.03

0.04

0.05

0.06

0.07

Time

lam

bda

VOICEVIDEOCALLDATA

Switch off 1 Node-B for about 9 hours

Energy saving= 37.5%


7 μCells with:R_μcells=100mPTX_μcells=2 W

Umbrella (Macro) Cell:R_Mcell≈265mPTX_Mcell=3.4 W

Hierarchical scenario


8:00 16:00 24:00 8:00 16:00 24:00 8:000

0.01

0.02

0.03

0.04

0.05

0.06

Time

lam

bd

a

VOICE

VIDEOCALL

DATA

λnight→0: Good for

office scenario


8:00 16:00 24:00 8:00 16:00 24:00 8:0010

-20

10-15

10-10

10-5

100

Time

Blo

ckin

g P

rob

abi

lity

VOICE

VIDEOCALL

DATA

8:00 16:00 24:00 8:00 16:00 24:00 8:0010

-4

10-3

10-2

10-1

Time

La

mb

da

µ M Mµ µ

-The Umbrella cell is always ON (day+night)- Switch off 2 Node-B (7µcells) for about 4 hours

Energy saving= 17%


Possible configurations

Manhattan configurations (linear)

(1,2)

(2,3)


Possible configurations

Hexagonal configurations (squared)

(3,4)

(8,9)


Switch off schemegeometry

(1,2)linear

(2,3) linear

(3,4) squared

(8,9) squared

Load ratio 2 3 4 9

Cell radius 2x 3x 2x 3x

PB[W] 5 18 5 18

Night zone 16h30m 14h40m 12h20m 7h

NodeB saving [%] 68.7 61.1 50.4 29.1

Network saving 34.3 40.7 37.8 25.9

Switching off more does not always mean saving more!


But, we have multiple operators

• Several competing mobile operators cover the same area with their equipment

• Networks are dimensioned over the peak hour traffic

• During low traffic periods the resources of one operator are sufficient to carry all the traffic

Make operators cooperate to reduce energy consumption


• In turn,– Switch off the network of one operator, when

traffic is low and the active operators can carry all the traffic

– Let users roam to other operators– Balance costs

Cooperation


• Two operators: A and B– No. of users NA and NB, with NB = NA and <1– Daily traffic profile fA (t) and fB (t), fB(t)= fA(t)

Example: 2 operators

fM

T/2 T=24h t

fA(t)

fB(t) fM


fM

T/2 T=24h t

fM/(1+)

fM/(1+)

Switch off time for B

Switch off A

MMBA f)f()f( f)(f)(f BBBB

MMBA f)f()f( f)(f)(f AAAA

B A


• Let pA and pB be the frequency with which A and B switch off

• Different strategies can be adopted for choosing the switching frequency

Switch-off policies


• Switch off the networks every other day, alternatively,

pA = pB

Balanced switch-off frequency


• Make the two networks carry the same roaming traffic (on average)

Balanced roaming cost

B

B

A

A

T

BB

T

AA dttfpdttfp

2/2/

)()(

traffic carried by B when A is off

traffic carried by A when B is off


• Make the two networks achieve the same energy saving

Balanced energy saving

)2()2( BBBAAA TCpTCp

switch off time for A

energy costfor A


• Two cost models– Constant: the fixed costs dominate, the

networks have the same energy cost regardless the number of subscribers

CB=CA

– Variable: the network energy cost is proportional to the number of subscribers

CB = CA

Balanced energy saving


Real traffic pattern

0

0.2

0.4

0.6

0.8

1

1.2

9:00 12:00 15:00 18:00 21:00 0:00 3:00 6:00 9:00

Tra

ffic

, f_

A(t

)

Time, t [h]


Constant cost model:Total saving

0

5

10

15

20

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

10

20

30

40

En

erg

y sa

vin

g [

cost

/day

]

En

erg

y sa

vin

g [

%]

Traffic ratio,

RoamingSaving

SwitchingMax

Saving can be huge!


Constant cost model:Roaming balance

0

5

10

15

20

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

10

20

30

40

En

erg

y sa

vin

g [

cost

/day

]

En

erg

y sa

vin

g [

%]

Traffic ratio,

TotalAB


Constant cost model:Switching balance

0

5

10

15

20

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

10

20

30

40

En

erg

y sa

vin

g [

cost

/day

]

En

erg

y sa

vin

g [

%]

Traffic ratio,

TotalAB


Variable cost model:Total saving

0

5

10

15

20

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

En

erg

y sa

vin

g [

cost

/day

]

Traffic ratio,

RoamingSaving

SwitchingMax

Different cost models lead to different

policies


Variable cost model:Total saving

5

10

15

20

25

30

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

En

erg

y sa

vin

g [

%]

Traffic ratio,

RoamingSaving

SwitchingMax

Different cost models lead to different

policies


Different QoS

• When operators guarantee different QoS levels, the network with the best QoS switches off only when the other operator can guarantee similar QoS

• This translates into a traffic reduction factor

MM f)f()(1 f)f()(1 AA

Same QoS Different QoS: QoS of A is tighter


2 Operators: Different QoS

0.4 0.5 0.6 0.7 0.8 0.9 1 0

5

10

15

20

25

0.1 0.2 0.3

Sav

ing

[%

]

QoS traffic reduction factor,

=0.25=0.50=0.75=1.00


2 Operators: Different QoS

20

21

22

23

24

0.2 0.4 0.6 0.8 1

Sw

itch

ing

tim

e

QoS traffic reduction factor,

On-Off

=0.25=0.50=0.75=1.00

6

7

8

0.2 0.4 0.6 0.8QoS traffic reduction factor,

Off-On

=0.25=0.50=0.75=1.00

1


Multiple Operators

• With more than 2 operators, the space of possible switch-off patterns explodes

• Different roaming schemes are possible, during the switch-off phase:– Roaming-to-One: Roaming traffic goes to the

operator which remains on all the time– Roaming-to-All: Roaming traffic is distributed to

active operators


Example: 4 Operators

• Let the number of users for operator i be proportional to i, with

a is the network unbalance– a=0: the networks carry the same traffic– a=1: network 1 has ¼ of the traffic of network 4

4

iaa)(1i


4 Operators: Increasing Pattern

20

25

30

35

40

0 0.2 0.4 0.6 0.8 1

Sav

ing

[%

]

Network unbalance, a

same cost - Allvar cost - All

same cost - Onevar cost - One

Under same cost, increasing pattern

is optimal

Roaming to all is more effective


4 Operators: Decreasing Pattern

20

25

30

35

40

0 0.2 0.4 0.6 0.8 1

Sav

ing

[%

]


same cost - Allvar cost - All

same cost - Onevar cost - One


4 Operators: Increasing Pattern

7

8

9

10

11

12

13

14

15

0 0.2 0.4 0.6 0.8 1

Off

tim

e


oper. 1 - Alloper. 2 - Alloper. 3 - All

oper. 1 - Oneoper. 2 - Oneoper. 3 - One


• Energy issues are crucial, even for networking– Design criteria must be changed – Energy consumption/wastage is a variable to

be taken into account in design and performance evaluation

– Future Internet design will have to cope with it– Virtual operators appear to be an interesting

option

Lessons


• A new attitude is needed– Consumers: awareness of the cost of

• Turn over of devices • Uncontrolled use of energy

– Manifacturers• Life cycle assessment

– Operators• Careful management of resources • Architectures

Lessons


• Governments and institutions will have to play a role in – Inducing new attitudes (e.g., education to an

aware use of resources)– Forcing new production models based on

products life cycle (e.g., responsibility for disposal, incentives to long lasting devices)

– Providing incentives for cooperation

Lessons


• M.Ajmone Marsan, L.Chiaraviglio, D.Ciullo, M.Meo, “Energy-Aware UMTS Access Networks”, W-GREEN 2008 - First International Workshop on Green Wireless, Lapland, Finland, September 2008

• M.Ajmone Marsan, L.Chiaraviglio. D. Ciullo, M.Meo, “Optimal Energy Savings in Cellular Access Networks”, GreenComm'09 - First International Workshop on Green Communications, Dresden, Germany, June 2009

• M.Ajmone Marsan, L.Chiaraviglio, D.Ciullo, M.Meo, “Energy-Efficient Management of UMTS Access Networks”, 21st International Teletraffic Congress (ITC 21), Paris, France, September 2009

• M. Ajmone Marsan, M. Meo, ”Energy Efficient Management of two Cellular Access Networks”, GreenMetrics 2009 Workshop, Seattle, WA, USA, June 2009

References


Thank you!


Questions?

Energy Efficient Wireless Internet Access

Technology

energy intensive

worldwide energy needs

california energy commission

wof electrical energy

wof primary energy source

department of energy

energyefficient way

data centers source