In IT energy costs, little steps can save a lot

26.10.2006
Using an automated system, Dick Hamann, the CIO at Seminole Community College in Sanford, Fla., has found a way to save enough money to cover a professor's annual salary: He turns off PCs at night at this 29,000-student college -- an effort that may trim enough from the school's power bills to save US$65,000.

Over the past six months, Hamann has been gradually rolling out automated shutdown technology from Persystent Technology Corp. in Tampa, Fla. His plan is to have the technology on 3,500 PCs by year's end.

And last week, Hamann began installing electric meters in his data center to learn how much it contributes to the college's annual $1.5 million energy bill. Once he knows that, he intends to go to his server vendor, Dell Inc., to see what it can do help him cut power usage. Reducing power costs "is what the college needs," he said.

IT managers are turning more attention to the cost of power as servers become more powerful and local utilities raise rates. According to Framingham, Mass.-based market research company IDC, the amount of money that users will spend worldwide on power and cooling servers will almost double from $26 billion last year to more than $44 billion by 2010.

But IT managers can't control what they can't easily measure, especially when it comes to vendor claims on server energy efficiency. With that issue in mind, a measurement protocol is being developed by the U.S. Environmental Protection Agency (EPA) and major IT vendors, with a final draft protocol for standardized testing of energy use due out by next week.

Server energy efficiency is very important to Paul Froutan, vice president of research and development at Rackspace Ltd., a managed hosting provider in San Antonio. Froutan conducts his own tests on server energy use and estimates that there may be as much as a 30 percent variation among similar server products in performance and energy use.

Rackspace has more than 20,000 servers, so even slight differences in energy efficiency could have a big effect on costs. For instance, if each server uses about 200 watts per hour -- and the energy consumption could be reduced by 2.5 percent, or 5 watts each -- that would save 100 kilowatts of energy. That's enough to power another 500 servers.

Workload isn't the only factor affecting a server's energy use. So does design. Some vendors put seven fans in their servers, others, five, said Froutan. "Do those five fans have to spin faster?" he asked.

Froutan was uncertain how much the efforts to devise a standardized measure of performance and energy use would help. The test workload might not exactly mimic the applications in his data center, he said. "Any data is better than no data," Froutan said, adding that he suspects he will still have to do is his own testing to supplement any vendor-supplied benchmark results.

Server makers now provide energy-use data, but they don't report that information uniformly. They may, for instance, offer an efficiency measurement of performance at 100 percent workload, which is not what customers typically use. The final draft protocol expected to be released by next week would measure performance per watt against a variety of workloads. The protocol's creators hope it will eventually be based on existing workload benchmarks developed by Standard Performance Evaluation Corp. and Transaction Processing Performance Council.

Bill Leo, CIO of Mercer Delta Consulting LLC in New York, said he doubts whether the effort to develop a benchmark will lead to energy improvements. "Unless they are going to be able to really compete with each other in this area, I don't know if it's going to have a whole lot of value," he said.

Leo also said users have to play a role. Vendors "need to take a real-world approach and meet with the users" and not develop a standard without user input, he said.

Dave Douglas, vice president for eco-responsibility at Sun Microsystems Inc., said the protocol will have to be paired with an existing benchmark. And while those benchmarks may not be real-world, most users have enough experience to know how a benchmark might compare with actual workloads.

In general, officials at Sun, IBM and Hewlett-Packard Co. all support the measurement protocol. But they said it's too early to know just when it will be ready for customer use.

Christian Belady, distinguished technologist at HP who has been directly involved in the effort, believes a uniform measurement protocol is inevitable because of customer concerns about saving power. "Ultimately, we believe it's going to be something that we have to do," he said.