Published on March 18, 2014
Energy use and the information economy Jonathan Koomey, Ph.D. Research Fellow, Steyer-Taylor Center for Energy Policy and Finance, Stanford University http://www.koomey.com Presented at The Physics of Sustainable Energy III University of California, Berkeley March 8, 2014 1
Defining “the Internet” 2
The big picture view 3Source: Ericsson and TeliaSonera (Malmodin and Lundén et al 2013) with support from CESC, KTH Sweden Key components • Data centers • Core network • Access networks • End-user communications equipment • End-user computing equipment Lots of complexity here!
The Internet is data… 4 Source: 1986 to 2007 adapted from Hilbert et. al. 2011; 2014 extrapolated using Cisco VNI data compiled at http://en.wikipedia.org/wiki/Internet_traffic Doubling time 1986 to 2014 = 3 years, doubling time 2000 to 2014 = 1.5 years. 1986 1993 2000 2007 2014E Mobile data Fixed Internet Voice
…but it’s also physical 5 Photo sources (clockwise from top left): Google. Flickr users Mr. T in DC, digger_90_tristar, geerlingguy, alachia, antonionicolaspina
∑: Growth in IT not necessarily = Growth in electricity use 6Copyright Jonathan G. Koomey 2013
Two opposing factors • Equivalently • So if Computations/Year goes up faster than Computations/kWh, then total kWh goes up! Copyright Jonathan G. Koomey 2013 7
What matters most… 8Source: Ericsson and TeliaSonera (Malmodin and Lundén et al 2013) with support from CESC, KTH Sweden. Data are for Sweden, circa 2010. These are key
The big three • End-user equipment • Data centers • Access networks 9
End-user Equipment 10
End user equipment 11 Computing – Desktops and local servers – Laptops – Tablets Communications – Phones – Wireless routers – Set-top boxes – Switches Display – Computer monitors – TVs (IP connected) Ultra low-power computing/sensors (small but growing)
Growing installed base of PCs worldwide 12 Source: IDC 2013 Vernon Turner
A key computing trend… 13Source: Koomey et. al. 2011 Energy efficiency of computing at peak performance up 100x every decade!
…led to the rise of tablets and mobile phones 14 Source: IDC (http://www.idc.com/getdoc.jsp?containerId=prUS24129713) Source: Hilbert and López 2012a and 2012b Tablet shipments = desktops in 2012!
Embedded emissions from manufacturing 15 Source: Koomey et. al. 2013 0% 20% 40% 60% 80% 100% Server (Mac Mini OS X server) Laptop computer (Macbook Pro 13") Smart Phone (iPhone5) NAND Flash memory - 1 GB Share of CO2 emissions Production Operation 0 200 400 600 800 1000 1200 Life Cycle CO2 emissions (kg) Percentage contributions Absolute emissions
Data centers 16
Kinds of data centers • Hyperscale (e.g., Google, Facebook, Microsoft, eBay, others) • Enterprise or “in-house” (vast majority) – Conventional – Internal cloud (similar to hyperscale) • Co-location (my facility, your IT) • High Performance Computing (special case– batch jobs, very high utilization) Copyright Jonathan G. Koomey 2013 17
18 Electricity Flows in Data Centers Copyright Jonathan G. Koomey 2013
Data centers used 1.3% of global electricity and 2% of US electricity in 2010* *For details see Koomey 2011 19Copyright Jonathan G. Koomey 2013
Data center electricity use worldwide 20 Source: Koomey 2011. Graph shows worldwide numbers. For the US, the range for data centers in 2010 was 1.7 to 2.2% of the total. N.B. Infrastructure in this slide refers to cooling, fans, pumps, and power distribution inside data centers.
Volume servers are dominant Copyright Jonathan G. Koomey 2013 21 Adapted from data in Koomey 2011
Server installed base 22 Source: IDC 2013 Vernon Turner
Why asset management is key Slide courtesy of Winston Saunders, Intel 23Copyright Jonathan G. Koomey 2013
Idle power improving: Server power curves (via Intel) • Usage Driven • Variable Utilization • Proportional Energy Use • Optimized Efficiency • Technology Scope: • CPU and Memory • Power Delivery, Fans, etc. • Instrumentation Approaching “Ideal” Server Behavior Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark* and MobileMark*, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. Configurations: Dual Socket Server. For full configuration information, please see backup. For more information go to http://www.intel.com/performance Xeon™ 5160 Xeon™ E5-2660 2012 2006 Data from spec.org Source: Winston Saunders, Intel 24Copyright Jonathan G. Koomey 2013
A key metric you will encounter • PUE = Power Utilization Effectiveness • PUE = • Measures infrastructure efficiency but says nothing about IT efficiency • Typical PUEs – Hyperscale/modular: 1.05 to 1.15 – New enterprise DCs: 1.25 to 1.5 – Existing enterprise DCs: 1.8 to 2.0 Copyright Jonathan G. Koomey 2013 25
A broader view • DC GHG emissions affected by 3 factors – Infrastructure efficiency (PUE) – IT efficiency – Emissions intensity of electricity • Best to focus on value, costs, and emissions per computation, not narrow efficiency metrics • More computations means higher total business value, lower costs per computation, and higher profits Copyright Jonathan G. Koomey 2013 26
PUE isn’t everything 27 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 Relative Data Center GHG Emissions (Baseline = 1) RelativeDataCenterEnergyUse(Baseline=1)       PUE=1.8, minimal IT efficiency PUE=1.5, minimal IT efficiency PUE=1.3 (free cooling, warm climate), minimal IT efficiency PUE=1.1 (free cooling, cool climate), minimal IT efficiency Decreasing electric power CO2 intensity Increasingoperationalenergyefficiency High energy, high carbon region High energy, low carbon region Low energy, low carbon region Baseline data center powered by coal: - Energy use = 92 GWh/yr - GHG emissions = 89 kt CO2e/yr PUE=1.1, maximal IT efficiency PUE=1.8, maximal IT efficiency [D] Coal (0.96 kt CO2e/GWh) [C] U.S. average electricity (0.6 kt CO2e/GWh) [B] Natural gas SOFC (0.35 kt CO2e/GWh) [A] Renewables (~0.02 kt CO2e/GWh) Source: Masanet et al. 2013
Many efficiency opportunities, particularly in IT equipment 28 Source: Masanet et al. 2011 Copyright Jonathan G. Koomey 2013
Energy-related advantages of hyperscale/cloud computing • Low PUE • Diversity of users • Economies of scale • Flexibility (because of abstraction/virtualization) • Easier provisioning for outside users ∑: Costs and energy use per computation much lower than conventional enterprise/colo installations Copyright Jonathan G. Koomey 2013 29
Annualized costs mostly IT, but infrastructure costs not trivial Copyright Jonathan G. Koomey 2013 30 Adapted from data in Koomey et al. 2007
Capital costs to build data center infrastructure • From Uptime Institute (Stanley and Schafer 2012) – Lowest $5M/MW of IT load – Mean $15M/MW of IT load – Highest: $25 M/MW of IT load • Modular/prefabricated solutions typically cheaper – Mean: < $10M/MW Copyright Jonathan G. Koomey 2013 31
What’s 1 W of IT savings worth? 32 Source of data: Koomey 2012. Infrastructure capital savings apply to new construction or existing facilities that are power/cooling constrained. Those savings total $8.6M/MW for cloud facilities and $15M/MW for others, from Uptime institute. PUE = 1.1, 1.5, and 1.8 for Cloud, New, and Existing data centers, respectively. Electricity price =$0.039/kWh for cloud facilities and $0.066/kWh for new/existing data centers. All costs in 2012 dollars. Copyright Jonathan G. Koomey 2013
Data center lessons • Biggest inefficiencies in enterprise data centers (cloud providers much better) • Just adopting best practices will save 80+% • Biggest impediments to efficiency are institutional, not technical • IT efficiency most important, followed by infrastructure efficiency and sourcing of low- carbon electricity (embedded emissions not so important) 33
DSE http://dse.ebay.com Copyright Jonathan G. Koomey 2013 34
Access networks 35
Access network bandwidth installed worldwide in 2010 36 Source: Hilbert and López 2012a and 2012b
Access network electricity use, Sweden 2010 37 Source: Ericsson and TeliaSonera (Malmodin and Lundén et al 2013) with support from CESC, KTH Sweden.
System effects of IT 38
Is the Internet an energy hog? 39 0% 1% 2% 3% 4% 5% Electricity consumption Energy consumption CO2 emissions GDP 1992-1996 1996-2000 Annual percentage growth N.B., the amount of data flowing through the Internet grew at about 100%/year in the late 1990s Source: Koomey et al. 2002.
System effects of IT • Dematerialization (move bits not atoms) – CDs vs downloads for music • Big systems optimization – Smart parking sensors reduce traffic • Enabling structural change – Flatter, nimbler organizations 40
Dematerialization:move bits not atoms 41 Source: Weber et. al. 2010 CO2 emissions for downloads and physical CDs -80% -40%
Big systems optimization: Smart parking 42Source: Mark Noworolski, Streetline Networks Motes use <400μW on Average. For LA, with 40,000 parking spots, that implies total mote power of about 15W. Mote technology is from Dust Networks
Structural change: Nimble organizations • IT enables business process redesign, improving efficiency across the board • Example: – Gave suppliers access to POS and inventory data, as well as company forecasts – Pioneered aggressive use of RFID – Improved the flow of supplies and finished goods – The result: Better coordination of suppliers with Walmart’s needs, plus much lower distribution costs 43
Suggested reading 44 Brynjolfsson, Erik, and Andrew McAffee. 2014. The Second Machine Age: Work, Progress, and Prosperity in a Time of Brilliant Technologies. New York, NY: W. W. Norton & Company. [http://amzn.to/1gYHEGk]
Key research issues 45
Key research issues • Need recent data on electricity use and potential savings • Need more system efficiency case studies • Need more and better automated reporting of – Energy use – User behavior • Average (fixed) vs marginal (variable) energy use – Most devices have high fixed energy use – Be careful to distinguish average vs marginal effects • Address rise of machine to machine communications 46
Conclusions • Popular preoccupation with electricity used by Internet-related systems is misplaced – Almost certainly <10% of total electricity, but not well characterized – End-user devices important, but most can’t be clearly allocated to “the Internet” • System effects potentially much more important than direct electricity use – IT affects the other 90% of electricity plus all the fuels • Updated data needed! 48
References • Brynjolfsson, Erik, and Lorin M. Hitt. 2000. "Beyond Computation: Information Technology, Organizational Transformation and Business Performance." Journal of Economic Perspectives. vol. 14, no. 4. Fall. pp. 23-48. • Hilbert, Martin, and Priscila López. 2011. "The World's Technological Capacity to Store, Communicate, and Compute Information." Science. vol. 332, no. 6025. April 1. pp. 60-65. • Hilbert, Martin, and Priscila López. 2012a. "Info Capacity| How to Measure the World’s Technological Capacity to Communicate, Store and Compute Information? Part I: Results and Scope." International Journal of Communication. vol. 6, pp. 956-979. [http://ijoc.org/ojs/index.php/ijoc/article/view/1562/742] • Hilbert, Martin, and Priscila López. 2012b. "Info Capacity| How to Measure the World’s Technological Capacity to Communicate, Store and Compute Information? Part II: Measurement Unit and Conclusions." International Journal of Communication. vol. 6, pp. 936-955. [http://ijoc.org/ojs/index.php/ijoc/article/view/1563/741] • Koomey et al. 2002. "Sorry, wrong number: The use and misuse of numerical facts in analysis and media reporting of energy issues." In Annual Review of Energy and the Environment 2002. Edited by R. H. Socolow, D. Anderson and J. Harte. Palo Alto, CA: Annual Reviews, Inc. (also LBNL- 50499). pp. 119-158. • Koomey, Jonathan, Kenneth G. Brill, W. Pitt Turner, John R. Stanley, and Bruce Taylor. 2007. A simple model for determining true total cost of ownership for data centers. Santa Fe, NM: The Uptime Institute. September. <http://www.uptimeinstitute.org/> • Koomey, Jonathan. 2008. "Worldwide electricity used in data centers." Environmental Research Letters. vol. 3, no. 034008. September 23. <http://stacks.iop.org/1748-9326/3/034008>. 50
References (continued) • Koomey, Jonathan G., Stephen Berard, Marla Sanchez, and Henry Wong. 2011. "Implications of Historical Trends in The Electrical Efficiency of Computing." IEEE Annals of the History of Computing. vol. 33, no. 3. July-September. pp. 2-10. [http://www.computer.org/csdl/mags/an/2011/03/man2011030046-abs.html] • Koomey, Jonathan. 2011. Growth in data center electricity use 2005 to 2010. Oakland, CA: Analytics Press. August 1. [http://www.analyticspress.com/datacenters.html] • Koomey, Jonathan G. 2012. The Economics of Green DRAM in Servers. Burlingame, CA: Analytics Press. November 2. [http://www.mediafire.com/view/uj8j4ibos8cd9j3/Full_report_for_econ_of_green_RAM-v7.pdf] • Koomey, Jonathan G., H. Scott Matthews, and Eric Williams. 2013. "Smart Everything: Will Intelligent Systems Reduce Resource Use?" The Annual Review of Environment and Resources.vol 38. October. pp. 311-343. [http://www.annualreviews.org/doi/abs/10.1146/annurev-environ-021512-110549]. • Masanet, Eric R., Richard E. Brown, Arman Shehabi, Jonathan G. Koomey, and Bruce Nordman. 2011. "Estimating the Energy Use and Efficiency Potential of U.S. Data Centers." Proceedings of the IEEE. vol. 99, no. 8. August. • Masanet, Eric, Arman Shehabi, and Jonathan Koomey. 2013. "Characteristics of Low-Carbon Data Centers." Nature Climate Change. July. Vol. 3, No. 7. pp. 627-630. [http://dx.doi.org/10.1038/nclimate1786] • Malmodin, Jens, Dag Lundén, Åsa Moberg, Greger Andersson, and Mikael Nilsson. 2013. "Life cycle assessment of ICT networks–carbon footprint and operational electricity use from the operator, national and subscriber perspective." Submitted to The Journal of Industrial Ecology. March 8. • Traub, Todd. 2012. "Wal-mart used technology to become supply chain leader." In Arkansas Business. July 2. [http://www.arkansasbusiness.com/article/85508/wal-mart-used-technology-to-become-supply-chain-leader] • Weber, Christopher, Jonathan G. Koomey, and Scott Matthews. 2010. "The Energy and Climate Change Impacts of Different Music Delivery Methods." The Journal of Industrial Ecology. vol. 14, no. 5. October. pp. 754–769. [http://dx.doi.org/10.1111/j.1530-9290.2010.00269.x] 51
Presentación que realice en el Evento Nacional de Gobierno Abierto, realizado los ...
In this presentation we will describe our experience developing with a highly dyna...
Presentation to the LITA Forum 7th November 2014 Albuquerque, NM
Un recorrido por los cambios que nos generará el wearabletech en el futuro
Um paralelo entre as novidades & mercado em Wearable Computing e Tecnologias Assis...
8.4.3 Energy Proportionality of ... Iyengar, Schmidt, Energy Consumption of Information ... point the laws of physics start to catch up and ...
... [UC BERKELEY -AFFILIATED] JOINT ... Although the majority of Koomey’s research focuses on energy ... an audacious plan to reduce the energy use of ...
... witnessed the dramatic increase of the use of information technology in ... Energy Physics: ... Eng. L. W. 31. Invited Talk @ UC IT ...