Reduce the Cost of Cooling Data Centers

As the price of running a data center continues to rise, researchers investigate novel ways to cut two of the top costs: administrative and energy expenses. Carnegie Mellon University created the Data Center Observatory to find techniques to trim the bills for both. Human costs (such as staffing, troubleshooting and maintenance) represent the largest ticket item, says Greg Ganger, professor at Carnegie Mellon and director of the Parallel Data Center, one of the university groups working on the observatory effort. Also, cooling costs add up quickly: In its lifetime, a data center can cost four to seven times more than the equipment’s original price tag, Ganger says.

In one part of the research, the observatory’s servers will be cooled with a new kind of system from American Power Conversion (APC), a major corporate partner for the project. Unlike a traditional cooling approach, where you’d cool the entire area where servers reside, APC’s system sections off large parts of the room with special plastic walls so that only the immediate area near the hot side of servers (where they vent out) needs cooling.

In order to reduce data center staffing expenses, the observatory will study exactly how data center employees spend their time—an area that’s not well understood today, Ganger says. Researchers will use the information to develop algorithms to automate more tasks.

The observatory’s researchers will also try to automate hardware troubleshooting, to save employees time. But this development job will be tough, says Granger, who characterizes it as a “long-range challenge.” It’s a hard problem to solve because of the sheer number of components and settings in a data center that can malfunction. Plus, computers aren’t inherently intuitive or creative about searching for sources of trouble, he says.