"When you think of it that way, it's a huge number," said Dave Sterlace, market development manager for the data center business at Thomas & Betts, the newest addition to the ABB family. Sterlace spewed a range of interesting numbers Tuesday afternoon at ABB Automation & Power World in Orlando, Fla., bringing into perspective just how massive power use is by the world's data centers — and ideas about how that power can be reined in.
By some estimates, power usage by data centers will double within five years. And data centers already using 15 to 30 times the power per square foot of an office building; some estimates put that at more like 50 to 100 times.
The largest data center in the world is a 100-MW facility near Chicago that belongs to Digital Realty, which hosts websites for other companies. In all of the Chicago metro area, only the O'Hare International Airport has a bigger power load. In the second half of 2012, Digital Realty added new capacity in metro New York City —an additional 30 MW. "They're big power users, they're going in extremely fast and they have an always-on mentality," Sterlace said.
That always-on mentality is essential for data centers. So while 2.5 is almost as small as 2, it's a huge problem when it's the average number of unplanned outages per year at U.S. data centers. "It's always got to be on," Sterlace says. "So 2.5 outages a year is a big problem."
All this said, there are trends in computing power at data centers that could have power usage — though still overwhelming — heading in the right direction. There has been a tremendous increase in computing efficiency, Sterlace notes. Server virtualization can equate to up to 80% fewer servers; blade servers increase the power density of racks; and the use of ARM and RISC processors can reduce power consumption by 90%.
Other trends also seem promising, Sterlace said. For example, higher voltages can make power distribution more efficient. You can increase power density using existing wiring; getting twice as much power through the same conductor equals savings in copper wiring.
There is much to be gained by the current movement toward the cloud as well. While current designs are based on hardware redundancy, the cloud is based on applications redundancy, leading to a less costly infrastructure. With the cloud, the use of data centers will be more likely able to "follow the moon" to save on standard energy costs — meaning that server usage could be switched to China, for example, while the Western part of the world is at higher usage levels. Scalable deployment is also key.
Sterlace also spoke of DC distribution, a topic that has already received considerable attention during the week's conference. But could it be the wave of the future? It just might be for leading-edge customers. "For the right person, it could be a really good solution," Sterlace says, explaining how DC can require less transformation and the transformation losses that go along with it.
Still, there are some hurdles that will need to be overcome for DC distribution to be viable on a larger scale. The National Electrical Code (NEC) doesn't yet recognize DC-only wiring, so you're not able to take full advantage of the higher power capacity of DC wiring. Grounding and arc flash standards are still in development, and there is still design work to be done.
"Some of the trends are very good ideas, but you have to determine for yourself which are relevant to your facility," Sterlace said. "Sifting through all the information is the hardest part."
