Energy Star rating can disguise data center inefficiency

The government’s Energy Star program took one on the chin earlier this year when it gave its seal of approval to a gasoline-powered alarm clock, a blatantly phony product that, along with 14 other fakes, won certification after being submitted as part of a covert investigation by the Government Accountability Office.

Fortunately, there has been no whiff of similar chicanery with a program launched in June that lets highly efficient data centers earn their own Energy Star label. For starters, a licensed professional must independently verify a data center’s energy performance and sign and seal the application that is sent to the Environmental Protection Agency for approval. That process isn't required for many consumer products in which manufacturers can self-report performance data.

Many praise the data center program for raising awareness and encouraging steps to control those facilities’ voracious and growing appetite for power. However, if EPA’s Energy Star is the only measure for energy efficiency success at your agency’s data centers, it’s probably best not to hoist any “Mission Accomplished” banners just yet.

That’s because the program relies heavily on a metric named Power Usage Effectiveness that, though certainly useful and endorsed by many energy efficiency mavens, has serious limitations as a tool for achieving greater levels of efficiency.

“PUE is a good starting point, but if you leave it at that, it doesn’t tell the whole story and becomes real dangerous,” said Greg Schulz, a consultant and author of "The Green and Virtual Data Center."

For EPA’s purposes, PUE is calculated by measuring how much of a building’s total energy is used by the IT equipment in the data center relative to the amount used by infrastructure such as lighting, power supplies and cooling equipment — PUE = total facility source energy/ IT source energy.

The closer a PUE is to 1, the more efficient that building is at providing a hospitable environment the IT equipment requires to operate. Industry estimates peg the average data center facility PUE at about 2.5. In the Energy Star rating for data centers program, EPA compares an applicant’s PUE with similar buildings nationwide and rates it on a 1-100 scale. A building must have a rating of 75, meaning it performs better than 75 percent of all similar buildings, or higher to earn the Energy Star label.

The problem is that a data center could have a great PUE and earn an Energy Star label but use twice as many computers and storage and network devices — and thus more energy — to handle a certain amount of data processing chores as does someone else’s data center, Schulz and others said. PUE tells you the efficiency of your building’s infrastructure but not your IT equipment, a critical distinction.

“If my servers are running at only 5 percent utilization, it doesn’t affect my PUE at all,” said David Cappuccio, vice president and chief of research for infrastructure and operations at Gartner.

EPA officials are aware of PUE’s limitations but chose it for the Energy Star program because it is accepted by the data center industry and easy to calculate. They agree that to work toward the next level of data center efficiency, you need to use a metric that connects energy consumption to the work output of the IT systems.

However, there is now no commonly accepted method for calculating that. EPA is participating in the industry discussions, and if appropriate metrics are developed, the agency will evaluate them for possible use in the Energy Star program, EPA officials said in a statement.

Gartner is among those who have taken a stab at developing such a metric. It’s Power to Performance Effectiveness rating allows managers to compare their IT equipment’s average performance against optimal maximum targets that they’ve set. The Green Grid, an industry group that helped promote PUE, is working on a similar metric named Data Center energy Productivity.

So is all the fuss even worth it? Absolutely. Cappuccio recently conducted a study that showed that trying to get a few extra years out of older servers might be more expensive than upgrading to more powerful and energy efficient newer equipment. An earlier upgrade might also let some agencies put off the need to build brand new data centers to keep up with growing data processing needs.

But you wouldn’t know any of that if PUE was all you looked at.