July 16th, 2002
Berkeley Lab Science Beat Berkeley Lab Science Beat
Data center energy use: truth versus myth
Berkeley Lab Science Beat

Lab website index

Lawrence Berkeley National Lab home page

Search Lab science articles archive
 Advanced Search  
Search Tips

Digital Economy's Demand for Steady Power Strains Utilities

Data Servers Crave Power: High-tech Electricity Needs Amplify Crisis

Net Blamed as Crisis Roils California

At the height of the electricity crisis of 2001, Californians were greeted with headlines like these over their morning coffee. One of the biggest misconceptions about the crisis was the role played by the energy use of computers and other internet-related hardware.

But early in 2001, research by Jon Koomey of Berkeley Lab's Environmental Energy Technologies Division (EETD) had showed that widely discussed estimates of the energy use of computer and networking-related hardware were exaggerated. Koomey is leader of EETD's End-Use Energy Forecasting Group; his work proved that this equipment used about three percent of U.S. electricity consumption -- a striking contrast to the 13 percent figure widely cited in the media.

New information on data centers
Questions persisted about the use of energy by a prominent type of facility that has become more common since the expansion of the internet as a commercial entity: the data center, also known as the web server farm. These buildings can house thousands of computers that store and transmit the data and web pages available on the Internet. Two recent developments at Berkeley Lab focus new attention on data centers.

Schematic layout of an internet data center (Sun Microsystems)

The first is a study by Jennifer Mitchell-Jackson, Koomey, and others, which concludes that the energy use of data centers is often overestimated. The second development is the announcement of a $500,000 grant to Berkeley Lab from the California Energy Commission, to study the stock of data centers in California, benchmark their energy use, and develop an R&D "roadmap" with the objective of reducing energy use in the state by 30 percent .

"Many reports of data center energy use are exaggerations," says Mitchell-Jackson. "They arise from a lack of measured data from operating data centers, inconsistent definitions of how to express the power consumption in these facilities, and use of rated or design power instead of actual power when estimating total consumption. Rated power is typically several times greater than actual power use."

"The question is an important one," says Koomey, "because overestimates of data center power use can leave utilities with expensive generation, transmission, and distribution capacity sitting idle."

Total electricity use about one-tenth of one percent
"The research found that in the U.S. there were about 9.5 million square feet of hosting-type data-center space in 2000. Our measurements suggest that these data centers have computer rooms that use an average of 50 watts per square foot or less," says Mitchell-Jackson.

She continues, "The total use of electricity by hosting-type data centers in 2000 was less than 500 megawatts of power, or 0.12 percent of the total electricity use in the United States in 2000." Total electricity used by these facilities is therefore small in the aggregate, although the clustering of data centers in certain regions may strain local electricity distribution and supplies.

Server-farm power density exaggerated
Over the past few years, utilities in California, New York, and Washington state have received requests for tens to hundreds of megawatts of electric capacity for new data centers. The requests created concern that these new power demands would overwhelm generating capacity in these states.

The power use of a data center is often measured by its power density -- the number of watts per square foot or square meter in the building. "We reviewed power bills of five data centers from across the country and found that the average computer power density is three to four times lower than the maximum power density that the facility was designed to accommodate," says Mitchell-Jackson. "Unfortunately, it is often this maximum power density that is cited in the media."

The team studied one data center in the San Francisco Bay Area in detail, measuring energy consumption of servers, power distribution units (PDUs), uninterruptible power supplies (UPS), air distribution, and other building loads. To express energy use accurately, Mitchell-Jackson developed a measurement called "total computer-room power density," a metric that is most representative of a data center's power needs because it includes power drawn by computers and all supporting equipment, including PDUs and UPSs; heating, ventilation, and air-conditioning systems; and lighting.

Measurements at the center in the Bay Area revealed that total computer-room power density at the facility was 33 watts per square foot. In the five facilities for which billing data were available, the figure was always below 40 watts per square foot.

"Our hope is that the data center industry and electric utilities will use this research to better estimate their energy needs, resulting in more efficient use of energy in these facilities, more reliable supplies to the data center operators, and more accurate planning for utilities," says Koomey.

The research was conducted by Jennifer Mitchell-Jackson, a graduate student in the Energy and Resources Group at the University of California at Berkeley, working with Jon Koomey and Bruce Nordman of Berkeley Lab's Environmental Energy Technologies Division. Their results will be reported in two publications, one currently under review by Energy (the international journal) and one that will appear in Resources, Conservation, and Recycling.

Additional information: