Vendors tussle over measuring server efficiency

The IT community called on manufacturers to agree on an energy efficiency metric for servers at a recent conference hosted by the Environmental Protection agency. But it's not likely that vendors will come to consensus any time soon.

SANTA CLARA, CALIF. -- Sun Microsystems held a joint conference with the Environmental Protection Agency recently to help tackle the growing crisis of power hungry data centers. Representatives from Intel, Advanced Micro Devices, Hewlett-Packard, IBM and others all said the same thing: We need to talk about metrics.

However, while everyone understands measuring server efficiency is important, every vendor is looking for the method that gives them an advantage.

For more information:

Water cooling takes center stage at Sun event

Making sense of metrics

"How do you measure energy efficiency? If you can't measure it, you can't manage it. These competing companies will have to come to a meeting of the minds," said conference host Jonathan Koomey, a scientist with Lawrence Berkeley National Laboratory and power expert at Stanford University. "No measurement is perfect -- you just have to come up with one that's the least wrong."

To its credit, Sun was the first major systems vendor to stick its neck out -- promoting power efficiency in its latest marketing push, hosting the event and by trotting out its own efficiency calculator it calls SWaP (space, watts and performance).

While its yet to be seen if there is more than marketing to the message, Sun's initiatives have brought the rest of the IT community together to discuss the issues.

Sun on metrics: SWaP

The SWaP equation takes the server's "performance" and divides that number by the amount of space the server takes up, multiplied by the amount of watts it uses.

Paul Durzan, director of product marketing for scalable systems at Sun said he hopes the industry responds to Sun's fairly simple equation, but the competition hasn't officially reacted to this offering.

"We've published fair rules for SWaP," Durzan said. "We don't want this just to be a Sun system."

Part of the problem with getting other people on board is that performance is a very subjective term. What's going on with your servers when you're not using them? What about applications that aren't making the most out of the processing power? How much does memory factor into power consumption? And most importantly, who cares how much power the server is consuming if you spend three times as much power cooling it?

At the end of the session, IT pros were asked to come up with some take away ideas for server vendors. Here are some of the ideas:
  • Research what the E.U. and Japan are doing about the issue.
  • Focus on server standards, and then take the data center environmentals into account.
  • Invite the interested parties conspicuously absent: Microsoft, Oracle and other software vendors and facility managers.
  • Assign a third party facilitator to bring other vendors to the table.
  • Identify popular benchmarks for core applications to have accurate data.
  • Come up with a new performance measurement, rather than processor performance, such as Web pages served or the number of users able to access an application at one time.

"I can look at all these metrics, but I don't see a big picture. Manufacturers will only publish metrics beneficial to them," said Christian Belady, a power expert at HP.

Belady found holes in SWaP. For example, a company can compensate for energy efficiency by packing components denser. He also poked at Sun's multi-core strategy, asking if a customer is going to care about 10 watts energy savings when the licensing cost is going up thousands of dollars for the extra cores [sic]?

But Belady admitted, "It's always easy to say 'No, that sucks' and not have a suggestion."

Energy Star could develop a standard

One solution to the stalemate might be a third party judge. The EPA's Energy Star program offers ratings for power efficiency for almost any electronic device imaginable. In fact, it currently offers rates for personal computers. But it hasn't tackled servers yet.

"At this point it's not on the agenda. The EPA is in the information gathering stage," said Koomey, who has worked extensively with the EPA on data center issues. "Computers come with all kinds of configuration options, but Energy Star has come up with ratings for them. It's not impossible,"

Andrew Fanara of the EPA's Energy Star product development team was less optimistic.

"It hasn't necessarily been discussed, whether or not [Energy Star] is appropriate for servers. You want to make sure the standard is relevant," Fanara said. "We want to come up with the right metric to managing energy efficiency, not the perfect measurement.

But while the EPA currently doesn't rank server power efficiency, it has adopted a third party measure that tracks the efficiency of server power supplies. Ecos Consulting, an energy efficiency consulting firm headquartered in Portland, Oregon, came up with the idea. A server power supply -- which is often made by a company other than the server manufacturer -- will qualify for this program if it operates at 80% or greater efficiency at 20%, 50% and 100% of rated load.

According to Ecos Consulting, most power supplies are only 60-70% efficient. But in order to reach that 80% efficiency level there is a price barrier of premium materials. But with an estimated savings of 301 kilowatt hours per server per year, experts think the power supply could be a first step.

"Power supplies are one area where there are server standards, and it has nothing to do with the processors. The easiest way to gain efficiency is to buy servers with the 80 plus rating on the power supply," Koomey said. "The people designing servers are focusing on the up front costs, but the cost of supporting less efficient power supplies is far more than the few dollars to improve the product.

Despite the seemingly simple answer of more efficient power supplies, HP and IBM representatives voiced doubts. For example, the cut throat pricing on low end servers might cancel out any efficiency benefits on the power supply if a company is just going to short change other components of the server. Also, dropping an 80 Plus power supply into every server isn't realistic considering quality testing the hardware would have to be extensive.

But despite differences of opinion, the conversation has begun.

Let us know what you think about the story; e-mail: Matt Stansberry, News Editor

Dig deeper on Data center backup power and power distribution

Pro+

Features

Enjoy the benefits of Pro+ membership, learn more and join.

0 comments

Oldest 

Forgot Password?

No problem! Submit your e-mail address below. We'll send you an email containing your password.

Your password has been sent to:

SearchWindowsServer

SearchEnterpriseLinux

SearchServerVirtualization

SearchCloudComputing

Close