Data center managers are looking at how a new breed of energy management software can help them make better decisions about which applications to virtualize or move to the cloud. But the success of these initiatives depends on their power-metering equipment, early adopters say.
Increasingly, vendors like SynapeSense, Power Assure, Modius, JouleX and others analyze data center power consumption data to make recommendations about how best to optimize server placements, both physical and virtual, to reduce data center costs, said Katie Broderick, senior research analyst at IDC.
“For a long time, the largest power and cooling solution was virtualization,” she said. “But if you’ve already done that -- now what?”
That played into NASA’s thinking last year when it deployed Power Assure energy management software across its 75 data centers. Working under Executive Order 13514 to reduce its carbon footprint, the space agency hoped that energy management would help it figure out its virtualization strategy, said Chris Kemp, NASA’s CTO of IT.
But several months into the project, the IT group leading the charge faces a catch-22: It doesn’t have enough meters in its data centers to gather the rich power consumption data it needs, nor does it have the money to invest in those meters.
“NASA’s under a lot of budget pressure,” Kemp said, “and sometimes it’s difficult to convince the IT organizations to make that up-front investment,”
Power metering equipment usually comes in the form of intelligent power distribution units (PDUs) -- sophisticated power strips with built-in meters. Whereas a generic PDU runs about $200, an intelligent PDU can easily cost $1,000 and up.
Newer servers, meanwhile, increasingly expose power consumption data through on-board management interfaces like Hewlett-Packard Co.’s Integrated Lights-Out, Dell’s Integrated Platform Management Interface and the like. But at NASA, there is a lot of older infrastructure. “We have to augment what is there,” Kemp said.
Complicating matters for NASA and many other large organizations is the fact that the IT department does not pay the power bill -- facilities does. Thus, the savings that could be achieved from better energy management will not accrue to IT. “The CIO needs to make an investment that will save NASA money, but that won’t show up on their bottom line, so they can’t get the return on investment,” Kemp said. Until then, Kemp said he is working with NASA’s CFO to come to work out an acceptable solution.
Avoiding data center buildouts
When the stars are aligned and all the necessary equipment is in place, energy management should have a tangible, positive effect on data center costs and operations.
The managed service arm of Fujitsu in the U.K. has used Emerson Network Power’s Aperture Configuration Manager to track the configuration and location of servers and data center infrastructure.
Since Fujitsu started using the tool in 2002, the company has has grown from about 45,000 to almost 300,000 square feet of data center space across nine data centers, housing tens of thousands of servers. Without Aperture, this kind of growth would have been impossible, said Chris Flanagan, Fujitsu data center operations development manager.
Last year, Fujitsu upgraded to the latest version of Aperture, and implemented workflow and capacity planning modules. The next step, Flanagan said, is to explore Aperture’s Integrated Resource Manager module. “The goal is to delay data center build-outs,” he said. “Building out a new data center is not cheap. The more accurate the data, the better the decision.”
To that end, Fujitsu will plug Aperture into a monitoring product such as Liebert Nform and SiteScan, or Avocent DSView3.
The data you have
But some vendors say that data centers with less-than-perfect instrumentation -- or none at all -- don’t have to sit on the sidelines. Sentilla, an energy management startup in Redwood City, Calif., claims it can estimate device energy consumption using so-called “virtual meters.”
“We talked to big companies that told us they want to profile their apps’ energy consumption,” said Joe Polastre, Sentilla CTO. “But they told us ‘I don’t want to install any new meters; it’s just not going to happen,’” he said.
For those shops, Sentilla Energy Manager uses a so-called Inference Engine to estimate energy consumption based on the equipment model, plus any power consumption data it can gather from branch circuit monitors, if available. Polastre said Sentilla’s virtual meters are at least 90% accurate.
And ultimately, some visibility is better than none, said NASA’s Kemp.
“I suspect where we’ll end up is a view of power consumption data with various resolutions based on the environment that we are talking about,” he said. “The CIOs that make the investments [in metering equipment] will be the best positioned to make decisions about which apps they should virtualize.”