e effi i energy efficiency in ict - oecdsoon 50% of operating expenses will be caused by power...
TRANSCRIPT
E Effi i Energy Efficiency in ICT
Prof. Dr.-Ing. Wolfgang Nebel
Page 122nd of May 2008 OFFIS – Institute for Information Technology
Energy Consumption, Climate Protection, Costs
Power Consumption of ICT (End Devices and Infrastructure) in Germany (Source: Cremer et al. 2003)
2001 38 TWh (8% f ti )2001: 38 TWh (8% of power consumption)Compared to 10.7 TWh generated from wind energy(Source: Bundesverband Windenergie)
2010: 55 TWh (11% of power consumption)
CO2-Emissions 2004 in GermanyICT: 28 Mio. t CO2 (due to power consumption)Compared to complete aircraft traffic of 22 Mio. t CO2p p 2
Increase especially in ICT-Infrastructure(UMTS-Network, Servers, Routers etc.)
The highest electricity consumers are data centersServers: 180 TWh in 2005 relate to 18 nuclear powerplants worldwide80% of all servers are located in data centers
Increase of Operating Expenses due to Energy CostsEnergy costs worldwide for servers in 2008: approx. 40 billion US$Soon 50% of operating expenses will be caused by power supply and cooling
Page 222nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel
Energy Consumption of ICT in Germany
ICT-infrastructure in private homes
Infrastructure mobile communication
Server, office
Audio stationary
Source: Borderstep 2007/Cremer et. al. 2003
TV
Page 322nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel
Worldwide Energy Consumption of Servers 2005 vs. 2000
3
2
2,5
1
1,5
Reference 2000:
0,5
1Reference 2000:
0
Volume
Rang
e
gh-E
ndTo
tal
AnzahlNumber
Source: Koomey, 2007
Vo
Mid-R
High T
Energie/ServerGesamtenergie
Energy/Server
Total Energy
Page 422nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel
The Economic Consequences
Source: IDC 2006
Page 522nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel
Problem Awareness of Data Center Operators
Power supply and cooling together: 74%
Page 622nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel
Energy Consumption in Enterprises (2,500 PCs)
1000000kWh
600000
800000
400000
600000
0
200000
0
ntro
lled
lized
PM
ched
off
Closing time
unco
n
cent
raliz
switc
h Closing timeWorking hours
Source: Gartner 2007
Page 722nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel
Core of the Problem: The Required Computing Performance
Source: tom‘s hardware guide
Page 822nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel
Cause: The Performance Barrier
Source: tom‘s hardware guide
Page 922nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel
Technical and Economic Challenges
Limited Run-TimeEnergy demand raises faster than battery capacityProblematic for mobile devices: E
nerg
y
Required by MobileDevices
Problematic for mobile devices:notebooks, mobile phones, PDAs, …
Added Costs
BatteryCapacity
Added CostsTypically relevant for devices for private use:desktop PCs, TVs, DVD players, …Of increasing importance in the commercial domain:
Generation
Of increasing importance in the commercial domain:e.g. work station PCs
Energy Supply and Disposal Problemgy pp y pIncreasingly problematicin data centersComplex air conditioning for 1:1p genergy disposal requiredSometimes, even the supplyof energy is a problem
Supply(electrical)
Disposal(thermal)
Page 1022nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel
Abstract Examination – Energy Efficiency
What is Energy Efficiency?ss efficient
notefficient
op
tim
ize
ffect
iven
e
C l i E ffi ie
Light BulbConclusion: Energy efficiency
can be achieved by:
• Optimized effectivenessEne g cons mption
EnergySaving Bulb
LEDIlluminant
um
pti
on
to u
sag
e
1Energy
always on
1 Energy
light dimmed, ifroom not used
1Energy Efficiency
light off, ifroom not used
• Energy consumptionproportional to usage
• High utilization
erg
y c
on
sup
ort
ion
al
t 1
EnergyEfficiency
Energy
EnergyEfficiency
Energy
en
ep
ro
efficient onlyat high utilization
alwaysefficient
Utilizationof the Room
1 Utilizationof the Room
1 Utilizationof the Room
1
Page 1122nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel
at high utilization efficient
Practical Implementation – Energy Efficiency in Mobile Phones
Optimize EffectivenessAdoption of low-power technologiesand design solutions in chip development
low VT Dual VT
Objective: minimizing steady-statepower dissipation
Raising Utilization Sleep States
high VT
Raising UtilizationBy means of load alignment
Energy Consumption Proportional to Usage DISPLAY
standby
Sleep States
Sophisticated power managementDisabling unused circuitsReducing performance during phases
ll d bg p g p
of lower utilizationObjective: Customizing the energy consumptionaccording to the respective usage scenario
VDD
Circuit
controlled byutilization
DVFS
Conclusion: Due to technical problems, vigoroususage of techniques to raise energy efficiency
Clk
Page 1222nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel
Practical Implementation – Energy Efficiency in PCs 1/3
Optimize EffectivenessOptimize EffectivenessTechnically feasible: low power CPU, low powergraphics card, 80% mains supply, thus no need for a fanToday instead of energy consumption rather performanceToday, instead of energy consumption rather performanceand pricing are important purchase criteriaOnly secondary effects (fan noise, high temperatures and short battery life time in notebooks) are reasons to purchase energy efficient systems) p gy y
Raising UtilizationMatching of the resources
„sending emails“ and „surfing“ needs no high-end PCMatching of the load
temporally - spatially
Energy Consumption Proportional to UsageTechnically feasible:
Shut down of individual components (hard disc display ) possibleShut down of individual components (hard disc, display, …) possibleLow power consumption during standby state
Controlled by software with a timeout-based power management
Page 1322nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel
Practical Implementation – Energy Efficiency in PCs 2/3
Problems of the Present Timeout Based Power Managements
Idea: After some idle-time, the useris considered to be absent-> turn off device (e.g. display)Error-prone (e.g. while user reads)Th h b l b t i The chosen balance between saving energyand annoyance often is to the disadvantageof energy saving
Solution: Intelligent Power Management
Intel06
Solution: Intelligent Power ManagementIdea: Adapt the power management to the pattern of useFor the first time, basic research was included in a product
Page 1422nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel
Practical Implementation – Energy Efficiency in PCs 3/3
Comparing Results of Adaptive vs. Timeout Based Power ManagementUse case: “Work of a traveling salesman”
Maximum at optimal power management83 i
Runtime of the notebookwithout PM: 5h
Maximum at optimal power management83min
54min62min adaptivetimeout
based
ng
er
nti
me
Due to breaks (phone call,reading of papers, …)max. extra time: 83min
21min
0min
lon
run
“adaptive” more energy efficient,with less false shut downs
mu
m u
sag
e
ap
top
nst
op
on
scalablean
no
yan
ceConclusion:
Energy efficiency is often a question of “financial sweeteners”
min
imb
att
ery
L
No
gy y qTechnical possibilities are available; but until now not sufficiently used
Page 1522nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel
Practical Implementation–Energy Efficiency in Data Centers 1/3
EffectivenessEnergy efficiency only if neutral to performance
Mains supply with high effectiveness (e.g. in Blade servers)
Smaller improvements arepossible, but one cannot
t j d t expect major advancements in the near future
Optimize UtilizationOptimize UtilizationAverage utilization oftoday’s servers (measuredapprox. 5,000 servers overapprox. 5,000 servers overa 6 month period)
Source: Google 2007
Page 1622nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel
Practical Implementation–Energy Efficiency in Data Centers 2/3
Optimize UtilizationAdditional strong dependencyon time and type of server
Optimization by consolidationImplementation of several servers
h d on one hardware resource
Realization by virtualizationl i t l tiseveral virtual operating
systems are independentlyexecuted on one singlehost systemhost system
Source: Vossel Solutions
Page 1722nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel
Practical Implementation–Energy Efficiency in Data Centers 3/3
Energy Consumption Proportional to Usage typical load
FrüherPast State-of-the-art Vision
1
Energy
Energy Efficiency1
Energy
EnergyEfficiency
1Energy
EnergyEfficiency
1
Energy
1
y
1
• Server without power management
• Power management inmodern CPUs result in
• Complete usage dependantenergy consumption
• Constant consumption• Energy efficiency only
depending onutilization
a minor utilizationdependant consumption
• in Blade servers loaddependant control ofmains supply
• Realizable by globalpower management inthe entire data center
mains supply
ConclusionRaising the load by consolidationNearly complete usage-proportional energy consumption is reachable with system wide
Research Projectat OFFIS
y p g p p gy p ypower management
Without penalty on performance With higher availability
Page 1822nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel
Practical Implementation – Possibilities by System Management
Hardware Resources virtualized serversVirtualized Data Center
R1…G1
kn
ow
n
G3
G2
ors
/o
ls
Rn R2R3Regarding their behavior, known and unknown virtualized servers
G5
un
kn
ow
nG6
G4
binding
monitors
mo
nit
oco
ntr
pla
ns
System Management
System Managementmonitors virtualized serversmonitors and controls the hardware resourcesplans binding
Optimi ation C ite ia
optimization criteria
1
energy
energy efficiency
Optimization CriteriaEnergy/Performancethermal behaviorh i
reduction ofinhomogeneous
aging
performance/energy
optimization
1
energyhomogeneous aging
prevention ofthermal
hot-spots
Page 1922nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel
g gp p
Power Consumption of German Data Centers
Source: Borderstep 2007
Scenario “state-of-the art”
Scenario “present efficiency trends”
Scenario “best-practice”
present development
Page 2022nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel
CO2 Emissions of German Data Centers
Source: Borderstep 2007
Scenario “present efficiency trends”
Scenario “state-of-the-art”
Scenario “best-practice”
present development
Page 2122nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel
Energy Costs of German Data Centers
Source: Borderstep 2007
Scenario “present efficiency trends”
Scenario “state-of-the-art”
Scenario “best-practice”
present development
Page 2222nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel
Practical Implementation – Example Projects
Strato’s CO2-free Data CenterffIncreased energy efficiency
Low power CPUs, optimized software (Sun Solaris 10)optimized building services and air conditioning
Energy supply by water powerEnergy supply by water power
The “green” Data Center of Host EuropeHighly available and energy efficient,Raised effectiveness of the supply components,cooling with outside air, …,30% lesser energy consumption
Bio-Gas Fuel Cell in T-Systems RZ-Munich“Our servers devour green stuff”250 kW of altogether 10 MW
Page 2322nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel
Finally …
Google 2005 1 cluster = 31,654 serversserved 40 Mio. requests/daycosts about 300 MWh/dayone request costs about 8 Wh
Second Life 2007approx. 2000 servers36,000 players simultaneously36,000 players simultaneouslyone avatar costs about 195 kWh/year or 129 kg CO2
Compare: a human exhales about 250 – 400 kg CO2
Thanks to Rolf Kersten: http://blogs.sun.com/rolfk/
Page 2422nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel
Conclusion
Technical problems already force manufacturers of mobile devices to implement radical new concepts. Increasing technical, economic and environmental interest in energy g , gyefficiency in data centers and ICT-infrastructures:
Potential and first technical alternatives are already available.Further innovative ideas have to be developed and evaluated.Holistic concepts have to be developed.Savings potential in 2010 more than 8 TWh / 5,4 Mio t CO2 / 1,1 Mrd. €
Very good prerequisites for applying new technologies, since y g p q pp y g g ,environmentally and economically reasonable,technically required and feasible.
ActionsRaise cost awareness in decision makers“Best practice” informationRoadmap to coordinate suppliers towards market launch Roadmap to coordinate suppliers towards market launch Flagship initiatives (improving the state-of-the-art)Energy-labels for the consumers market
Page 2522nd of May 2008OFFIS – Institute for Information Technology
Prof. Dr.-Ing. Wolfgang Nebel