Metric may point users to more energy-efficient servers

You can walk into the appliance section of any department store and quickly compare the energy consumption ratings of different refrigerators. But try to do that with servers, and you're out of luck.

By year's end, that may change. Standard Performance Evaluation Corp. (SPEC), a nonprofit organization based in the U.S., is expected to publish late this year a power utilization benchmark that's designed to enable IT managers to compare the energy efficiency of servers from vendor to vendor.

SPEC benchmarks are widely used by hardware vendors to measure the maximum possible performance of servers and usually are touted as part of the chest-thumping marketing exercises that follow new-product releases. But the new benchmark, which has been in the works for the past 18 months, will shed light on an area that has been difficult for hardware buyers to measure.

Klaus-Dieter Lange, a senior performance engineer at Hewlett-Packard who heads the SPEC subcommittee that is finalizing the benchmark, thinks the power usage test will drive vendors to improve the energy efficiency of their server designs.

Lange said that he expects the benchmark to be adopted by hardware makers shortly after its release, and he predicted that users will see "huge differences" among products in power usage.

Last week, the U.S. Environmental Protection Agency released the findings of a study on data center energy usage. The EPA's report included a stark warning: Unless IT managers adopt best practices in power management and make installing more energy-efficient technologies a priority, power consumption by data centers in the U.S. will double over the next five years.

Jonathan Koomey, a staff scientist at Lawrence Berkeley National Laboratory who has been advising the EPA on data center energy issues, said he thinks SPEC's benchmark "could have a very substantial impact" in the server market.

Currently, there's no way for IT managers to choose among servers based on the energy efficiency of the systems, Koomey said, adding that the new benchmark will allow "sensible decision-making." He said the benchmark may also lead to improvements of more than 30% in the power efficiency of servers as vendors respond with lower-power chips, better power supplies and other improvements.

But whether IT managers take advantage of the benchmark data is another matter. In many cases, IT budgets are separate from the budgets for running corporate facilities. "A lot of companies haven't worked out the incentive problems," Koomey said.

Ken Brill, founder and executive director of The Uptime Institute, said enterprise IT managers should pay attention to the EPA's report and take action if necessary. But he added that change may not come until businesses' chief financial officers take notice of energy bills and seek to have them reduced.

"IT can either deal with this, or the CFO is going to deal with it," said Brill. "I would prefer that IT deal with it."

In addition to HP, the companies represented on the SPEC committee that is drafting the new benchmark include IBM, Sun Microsystems, Dell, Advanced Micro Devices Inc. and Fujitsu Siemens Computers.

Lange said the initial version of the benchmark will use a Java-based application to measure energy usage not only at the maximum processing workload, but at lower performance levels as well. The benchmark will allow comparisons across a variety of microprocessor platforms, including x86, Itanium, UltraSparc and Power.

Before a vendor's benchmark data can be published, it first will have to be submitted to SPEC for review by the nonprofit organization and rival hardware makers. If no issues are raised during the two-week review period, the data will be cleared for public release, Lange said.