The U.S. Environmental Protection Agency expects to introduce its first Energy Star rating for servers by the end of the year, although a more comprehensive system that measures actual workloads will take longer to develop.

The Energy Star program is designed to make it easier for customers to identify the most energy-efficient products on the market. It is already offered for more than 50 kinds of products, including desktop PCs, monitors, ceiling fans and even windows, but the rating system for servers has been much harder to develop.

"This server program is one of the most complicated we've tried to deal with," said Arthur Howard, an associate with ICF International, which provides technical consulting to the EPA for its Energy Star programs.

That's partly because servers are used for so many types of work. Server makers say a benchmark test that measures power efficiency using one type of workload, such as file serving, won't provide meaningful results for customers that use the server for a different task, such as online transaction processing.

The EPA has been gathering input from server makers and other stakeholders for about a year. It quickly determined they would not be able to agree on a way to measure the "useful work" a server can perform with a given amount of power, said Andrew Fanara, who heads the Energy Star product development team.

"We all knew that in the long run, the most intellectually satisfying approach would be to marry energy consumption with work completed, yet admittedly we are not quite there yet in devising that holistic metric," he said in a recent interview.

The EPA hopes to use benchmark tests developed by a nonprofit group called the Standard Performance Evaluation Corp., or SPEC, but so far that group has published only one test for measuring power efficiency, based on a Java workload. It expects to publish several others but has not said when.

The EPA decided to sidestep the issue and come up with an initial, "Tier 1" rating that addresses two key areas it thinks can be measured. One is the efficiency of a server's power supply, which can be measured at various load levels. The other is how much power a server consumes at idle, when it isn't doing any real work.

The EPA will meet with stakeholders at Microsoft's campus in Redmond, Washington, next week to try to hash out definitions for a second draft specification, which could be ready for use by the end of the year, Fanara said. He acknowledged that it will be something of a stop-gap measure.

"If we were to develop this Tier 1 and then leave it in place indefinitely, I think it would not sufficiently recognize the most efficient products that emerged over time," he said. The idea will be to leave it in place for a limited period, while "on a slower track" the EPA figures out the type of "holistic measure" that it uses for other products, he said.

But the EPA has its work cut out for it even on the Tier 1 spec. It needs to work out which types of servers it will cover -- industry-standard servers with one to four processors are likely to be included -- and even the meaning of "idle."

"What's the definition of idle?" asked Mark Monroe, Sun Microsystems' director of sustainable computing. "How much of the OS has to be alive? Does idle mean it's responding to wake-on-LAN-type things, or is it truly in a bare minimum state?"

Like other vendors, Sun says it supports the Energy Star program in principle. But Monroe said the SPEC benchmarks being considered are untested and don't measure enough different kinds of workloads.

Some observers blame the server makers for the holdup. With customers paying more attention to power efficiency, vendors don't want to risk signing off on a specification that will make their products look bad, said Jon Peddie, president of Jon Peddie Research. "I think it is totally a political problem," he said.

He also questioned the usefulness of measuring efficiency at idle, especially as more companies are using virtualization to improve the utilization rates of servers. "It's like measuring miles per gallon when your car is sitting on the drive at home with the engine running," he said. "Who can use that?"

The difference, according to Howard, the EPA consultant, is that people don't leave their car engines running all night. But many data centers don't turn their servers off when they are not being used, because they are worried they won't come on again when needed.

Brad Brech, a distinguished engineer with IBM, said the measurements for the Tier 1 spec, though imperfect, will be a step in the right direction.

"I do believe we need more holistic measures as we go forward," he said, "but as with everything, having a good starting point gets the ball rolling."

The EPA is also working on an Energy Star rating for data centers, and Fanara said it will start to tackle storage equipment in the fourth quarter.