"Unless energy efficiency is improved beyond current trends, the federal government's electricity cost for servers and data centers could be nearly $740 million annually by 2011, with a peak load of approximately 1.2 GW."
That was the grim forecast of a 2007 EPA report to Congress on server and data center energy efficiency. And that's just for the federal government. By 2011, the national figures forecast nearly doubled the 2006 levels of estimated server and data center energy consumption, from about 61 billion kilowatt-hours (kWh) to more than 100 billion kWh.
As part of its recommendations to Congress, EPA called for "an objective, credible energy performance rating system for data centers" to be created, and that put the fire under a data-gathering initiative which culminated in the release this past June of the Energy Star label for data centers.
At press time, five data centers have earned the Energy Star label and around 100 more are benchmarking their energy performance in Portfolio Manager, the online Energy Star benchmarking tool used to calculate Energy Star performance.
Since its inception 18 years ago, the Energy Star label has become a well-known benchmark of energy efficiency. Now the same stamp of performance is available to data centers. But the new label has a few idiosyncrasies due to the special demands of data centers. Those quirks — coupled with the imperative to maintain reliability — make lessons from the first Energy Star labeled data centers useful to all facility managers responsible for data centers.
The first challenge in creating the energy performance rating system EPA called for in 2007 was that there was no existing data set for the agency to use. Normally, the Energy Star label is based on a data sample of around 100 to 400 facilities pulled from the U.S. Energy Information Administration's Commercial Buildings Energy Consumption Survey (CBECS). However, CBECS does not currently gather data on data centers.
Energy Star grades on a 1 to 100 rating scale for energy efficiency. A building with a rating of 75 is doing better than 75 percent of similar buildings. For this scale to work, at least 100 data points are needed and in late 2007 EPA began recruiting volunteers to provide at least 11 months of energy consumption data as well as provide operating characteristics data. By mid 2009, EPA had gathered data from 120 data centers of all types, sizes, and tier levels from around the country.
Then came dealing with the myriad peculiarities of data centers, which just might be the least homogenous facility type there is. Normally, Energy Star uses Energy Use Intensity as the dependent variable for its rating models, but this was not considered the best choice for measurement of data centers. Instead, EPA used Power Usage Effectiveness (PUE). PUE is the total energy used by a data center divided by the energy consumed by the IT load. The closer to 1, the more efficient a data center is, says Alyssa Quarforth, national program manager, Energy Star Commercial Properties, EPA. "Most of the industry has a PUE greater than 2.0," she says, adding that 3.0 and above is really inefficient. "But the leaders in the industry are around a 1.3 or 1.5."
Using PUE brings its own challenges. "It's unique from a building perspective in that the major portion of the energy equation on data centers is the IT load," says Alan Konieczynski, realty services asset manager for the Citi Midwest Data Center, one of the first data centers to earn the Energy Star. "And that portion of the energy equation is not factored into the Energy Star itself." The PUE calculation used for Energy Star is as much a function of the amount of critical load in the data center as it is a measure of the efficiency of the electrical and HVAC equipment supporting that load, he says, leaving a question of how to approach new buildings. A newer Citi data center facility which has a lower IT load than the Midwest Data Center has a much higher PUE. "Their Energy Star rating is lower than ours even though the design parameters and equipment are virtually identical to what we have," Konieczynski says.
Another challenge is that larger buildings, with a larger base amount of energy needed just to open the doors for business, may again have a PUE that's not so great when the facility is brand new with a light IT load, says BJ Butler, Citi realty services senior vice president for Citi's North American data centers.
Just where to measure the IT energy use was tricky in itself. It was decided to go with the output of the UPS, because this is what the majority of the data center industry can do at this time. In a perfect world, the ideal measure of IT energy consumption would be made at each server, but that sort of metering is not feasible at this point. Being able to express the amount of work that servers can do for the energy they consume is also attractive, and elusive.
"The industry is trying to get to a place where they can talk about compute power per unit of energy, but there's so many platforms there," says Daniel Gaffney, vice president and director of the critical infrastructure group with BNY Mellon, which is also a member of The Green Grid. "This is a very basic platform that everybody understands and everybody can then trend and over time become more efficient. Because obviously the goal is to become more efficient, reduce your carbon footprint and manage costs to the best extent that you can."
Redundant Redundancy?One tension in data centers is the aversion to downtime, which varies by how critical the mission of the enterprise is perceived to be. "Internet facilities are your biggest data centers and they plan on a failure rate," says BJ Butler, Citi realty services senior vice president for North American data centers. Depending on the enterprise, some amount of downtime may not harm their ability to deliver within customer expectations. Not so in Butler's world. "In the financial sector, a data center failure could be the difference between your ATM card working or not." To ensure uptime, redundant systems are established within data centers, which leads to greater energy consumption. "Typically, large modern data centers serving financial institutions will be designed with a full 2N redundancy with multiple electrical paths around each piece of equipment for concurrent maintainability and fault tolerance," says Alan Konieczynski, realty services asset manager for the Citi Midwest Data Center. Data Centers Too Conservative?Right now, redundancy or tier level is not factored into the Energy Star equation for the data center label. In establishing rating models, Energy Star normalizes for the large, unavoidable, energy use basic to the business. For example, for grocery stores, it's refrigeration power. In data centers, it's IT load. "We normalize for IT load," says Alyssa Quarforth, national program manager, Energy Star Commercial Properties, EPA. And while facility managers might not be in control of the ITequipment, they are in control of the other energy consumed. "So do you really need all of that energy to carry out your mission at the end of the day?" In other words, while the number of servers needed may be immutable, the number of chillers might not be. In crunching the raw data to make the model, EPA considered a range of variables: redundancy, tier level, number of racks, building type, etc. "None of those came out to be statistically significant, so they were excluded from the final model," says Quarforth. "Specifically for tier levels, it was found that data centers may have unnecessarily high tier levels, so normalizing for tier level could provide a disincentive for an energy efficient design." While the industry would love tier level to be considered a business constraint as well, energy efficiency and uptime goals don't have to be mutually exclusive. Both just have to be incorporated into the planning of a data center early in the process. "Most engineering firms and organizations go in with an idea of how energy efficient they want to be right from the start," says Daniel Gaffney, vice president and director of the critical infrastructure group with BNY Mellon. "But there's a tradeoff between energy efficiency and the amount of infrastructure you have to put in for increased redundancy. So you really have to look at it as a risk, try to fine tune things and work closely with your engineers when you're doing the basis of design." — Naomi Millán |
New Energy Star Rating System An Accessible Benchmark for Industry
First to Earn Data Center Energy Star Share Their Strategies
