1. Environmental Energy Technologies Internetdatatalk010507.ppt Electricity used by the Internet: How an urban legend became conventional wisdom Jonathan Koomey, Ph.D. Lawrence Berkeley National Laboratory JGKoomey@lbl.gov, 510/486-5974, http://enduse.lbl.gov/ For more details on the analysis discussed here, go to http://N4E.LBL.gov Download: http://enduse.lbl.gov/shareddata/internetdatatalk010507.ppt Presented at the EIA Spring conference, Environmental Issues Council Meeting May 7, 2001

2. Environmental Energy Technologies Internetdatatalk010507.ppt Background  The Internet Begins with Coal, by Mark Mills, was published in Forbes in May 1999.  Results were widely cited and have become conventional wisdom — Internet = 8% of all electricity use — All office equipment = 13% of all electricity use — Total projected to grow to 50% of all electricity use in 10-20 years

3. Environmental Energy Technologies Internetdatatalk010507.ppt Mills’ broader claims  He says electricity demand growth will accelerate because of increased numbers and use of computers and office equipment (he equates importance in $ to importance in kWh)  He says alternative sources cannot meet this increased demand, only coal can.  He says that any attempt to constrain demand will kill the information economy.

4. Environmental Energy Technologies Internetdatatalk010507.ppt The reality  The importance of information technology to the economy does NOT imply that it must use a lot of electric power.  Actual total for all office equipment is 3% of all electricity use, Internet electricity (using Mills’ own definitions) is only 1%.  Alternative sources are growing quickly and becoming an important industry (e.g., wind).

5. Environmental Energy Technologies Internetdatatalk010507.ppt Recent developments  Mills’ claims are becoming less widely believed, but many people still cite them, and some people still pay many $ to hear them. — Many institutions have backed away from using Mills’ numbers. — I and my colleagues have educated dozens of reporters on this topic.

6. Environmental Energy Technologies Internetdatatalk010507.ppt Mills’ analysis was flawed  Bad data: 1-2 kW PCs, 500 kW central offices, 250 kW mainframes, 1 kW routers in LANs.  Bad boundary definitions: Just what part of a PC or a server is “associated with the internet”, anyway?  Overlooked systemic effects: information tech. can affect the efficiency of many processes.

7. Environmental Energy Technologies Internetdatatalk010507.ppt Mills’ electricity “used by the internet”, corrected by LBNL TWh per year

8. Environmental Energy Technologies Internetdatatalk010507.ppt Data needs  Equipment sales  Power use (W) by operating mode  Usage (Hours) by operating modes  Equipment stocks (estimate using sales and lifetimes, and/or assess using surveys)  Materials use, reuse, and recycling  Estimate energy use and emissions  Systemic effects (e.g., telecommuting, e- commerce)

9. Environmental Energy Technologies Internetdatatalk010507.ppt Latest estimate of office equipment electricity use by sector in 1999 Residential 12% Industrial 13% Network 4% Commercial 71% Total = 74 TWh http://enduse.lbl.gov/Projects/InfoTech.html

10. Environmental Energy Technologies Internetdatatalk010507.ppt Comparison of Estimates of Energy Use for Commercial Office Equipment in 1999 Printer Desktop/Portable/Server Computer Display/Terminal Mainframe/Minicomputer Copier/Fax Annual Energy Use (TWh) http://enduse.lbl.gov/Projects/InfoTech.html

11. Environmental Energy Technologies Internetdatatalk010507.ppt A key uncertainty: Data centers/server hotels  Some utilities receiving requests for tens to hundreds of MW of power from proposed data centers  Estimates for power use almost certainly too large — One facility maximum = 90 W/sf, actual <40W/sf. — Another facility claimed 65 W/sf, but floor area defined incorrectly to exclude aisles and other common areas, leading to an overestimate of power use (this definition is critical) — Still another facility took direct server load and multiplied by three (!) to account for cooling, lighting, and other uses (implies a COP of about 0.5, 3-4 times too low)

12. Environmental Energy Technologies Internetdatatalk010507.ppt local distribution lines to the building, 480 V HVAC system lights, office space, etc. UPSPDUcomputer racks backup diesel generators Electricity Flows in Data Centers computer equipment uninterruptible load

13. Environmental Energy Technologies Internetdatatalk010507.ppt Defining metrics for data centers Power Loads (in watts) Footprint (in square feet) computer equipment (e.g., server) one rack (e.g., rack of servers) all racks all core data center space (net floor space) whole building (gross floor space) computers HVAC (in data center and for whole facility) lights (in data center and for whole facility) mechanical equipment excluding HVAC (e.g., UPS, PDU) single data center floor within a multipurpose office building other (e.g., copiers and faxs in office space) Utilities Data center designers, media total power load W/ft 2

14. Environmental Energy Technologies Internetdatatalk010507.ppt Upper bound estimate to electricity used by data centers in the U.S (1) Floor area taken from Juarez, Richard A. 2000. Virtual Bricks II: Virtual Econ 101 Update: A Comprehensive Guide for Understanding eCommerce Infrastructure Evolution and Convergence. Boston, MA: FleetBoston Robertson Stephens Inc. May. There is big uncertainty in the floor area number (could be as much as a factor of two too high). (2) Power density is an upper bound based on recent discussions with experts. Ignores the possibility that server loads are shifted from existing installations and not new loads. (3) Electricity use calculated assuming 8760 hours per year operation, flat load curve. (4) Total U.S. electricity use taken from EIA’s Annual Energy Outlook 2000. Contact: JGKoomey@lbl.gov, 510/486-5974.

15. Environmental Energy Technologies Internetdatatalk010507.ppt Possible systemic effects of information technology (IT)  Commercial floor space — Reduced office space — Reduced warehouse space  Changes in travel patterns  More efficient use of energy and materials  IT + utility deregulation = energy service markets for big service providers

16. Environmental Energy Technologies Internetdatatalk010507.ppt Something unusual may be going on Source: Joe Romm, Center for Energy and Climate Solutions

17. Environmental Energy Technologies Internetdatatalk010507.ppt Conclusions  Misinformation seems to spread more quickly than truth.  Bad data = wrong decisions  Computers and other office equipment are NOT the cause of the California power crisis.  Electricity used by computers and network equipment, including telecommunications and manufacturing energy = 3% of U.S. electricity use U.S., not 13% as Mills implies.

18. Environmental Energy Technologies Internetdatatalk010507.ppt Conclusions (p.2)  No evidence that a Palm VII uses as much as a refrigerator (Mills has ignored eight data requests from me on this issue).  Electricity used by data centers is relatively small in the aggregate; in certain cities such centers can be important contributors to load.

19. Environmental Energy Technologies Internetdatatalk010507.ppt Conclusions (p.3)  Understanding the systemic effects of information technology is critically important (CECS, CMU, UCB, and others)  Please spread the word!  Credible data are urgently needed. Join our network for energy, environment, efficiency, and the information economy (N4E) at http://n4e.lbl.gov

20. Environmental Energy Technologies Internetdatatalk010507.ppt If you really want to get the numbers right http://www.numbersintoknowledge.com New book, just released!