Recycled HPC system to be used to develop sustainable power consumption

The biggest challenge in computing today, some experts say, is not processing power, but power consumption. In 2007, the Environmental Protection Agency forecasted that as of 2011, data centers will be responsible for 2 percent of all power consumption in the U.S., and some predictions foresee those levels rising to almost 6 percent by 2020. Finally, there are numerous anecdotes about power demands caused by data centers, including partial brownouts when supercomputers are switched on and new data centers having to be moved to where cheap hydro-power is available, such as the Columbia River Gorge.

Clearly, power consumption is not only an environmental concern, but also a productivity and security issue. If high-performance computing (HPC) centers are going to be able to run larger simulations and process more and more data, they must find a way to decrease their facilities' drain on the power grid.

To help understand and reduce power consumption, the Georgia Institute of Technology has launched Green IT. The effort considers power consumption across the entire "energy stack," ranging from the power consumed by modern multi-core platforms, to the board and rack levels, to the entire data center. Corralling expertise from Georgia Tech's College of Computing, College of Engineering and Office of Information Technology, the consortium is a multidisciplinary effort that looks at how to build large-scale systems that use less power. The goal is to better understand where and how power is used, and to make it possible to coordinate power usage across different data center components, such as the cooling and the IT infrastructure.

"With experts from computer science looking at systems management, cloud computing and virtualization, and electrical engineers investigating chip design along with mechanical engineers working on cooling technologies, Georgia Tech is in a great position to help solve the power consumption problem," said Karsten Schwan, a professor in Georgia Tech's College of Computing.

Often, research efforts like these must use simulated machines, with heaters substituting for computers; but the Green IT group will be using a large-scale commodity system, a 1,000-node IBM BladeCenter, to conduct its investigations. The system was previously used by the Center for the Study of Systems Biology.

"Rather than junking the old machine, Georgia Tech decided that we could recycle it and use it for energy-efficient IT research along with a host of other uses," said Schwan.

The GreenIT effort is led by Sudhakar Yalamanchili in Electrical and Computer Engineering and includes the following faculty members: Ada Gavrilovska, Ron Hutchins, Yogendra Joshi, Hyesoon Kim, Hsien-Hsin Lee, Saibal Mukhopadhyay, Santosh Pande, Calton Pu, Karsten Schwan, Madhavan Swaminathan, Yorai Wardi, Marilyn Wolf and Jun Xu.

This week, Georgia Tech is showcasing research activities in high-performance computing and the computational sciences at SC09. The conference takes place at the Oregon Convention Center in Portland, Oregon, Nov. 14-20. Researchers and staff will be on hand at Booth 132 to demonstrate and discuss Georgia Tech's latest research and activities in the field.