Moving research to energy efficient community cluster saves time and money
Electrical engineering Professor Scott Sudhoff’s research looks, in part, for ways to optimize electric motors and generators. It is fitting then that Purdue’s community cluster supercomputers provided an opportunity to improve the energy efficiency of his research computing, while also giving Sudhoff a more powerful, more reliable tool for his work.
Sudhoff’s recent shift to the Rossmann community cluster enabled the retirement of a smaller cluster built for his lab in 2003 and housed in the Electrical Engineering Building. The retirement of the 2003 system should save Purdue more than $66,000 in power and cooling costs over the next four years. Sudhoff’s switch to Rossmann is part of Purdue’s ongoing data center consolidation project, which is saving the West Lafayette campus nearly $200,000 per year so far.
The addition of Sudhoff’s research to the Rossmann cluster has a negligible impact on the cluster’s power and cooling demands. Meanwhile, Rossmann’s components are more energy efficient than the decommissioned 2003 cluster. Rosmann was built in the fall of 2010.
Rossmann is housed in the Mathematical Sciences Building data center where ITaP’s Rosen Center for Advanced Computing operates four community clusters built since 2009 for use by faculty researchers and their students. The community clusters are serving hundreds of users in engineering, biology, chemistry, physics, agriculture, climate science, communications, computer science and other fields.
Besides energy efficiency, Sudhoff sees another advantage in shifting his research from an individual lab cluster to a community cluster: his lab can now concentrate on research rather than system administration. That’s because the Rosen Center’s expert staff does the work of operating and maintaining the community clusters, including security, software installation and user support.
“As the cluster in EE 139 aged, it developed maintenance issues,” Sudhoff says. “It started to seriously hamper our ability to do research. As we looked for alternatives, the Community Cluster Program was appealing in that it would seem to be an effective use of resources — and it gets us out of the computer maintenance business.”
The data center consolidation program — sponsored by the Office of the University Chief Information Officer and the Office of the Executive Vice President and Treasurer — offers loans for departments or faculty members who want to move to new technology in a consolidated data center. The loans, paid back with savings from reduced energy and infrastructure costs, fund replacement of outdated systems.
Faculty and administrative staff managing their own servers or data centers should contact John L. Smith, virtualization manager for ITaP’s IT Systems and Operations group, if they would like to participate in the program. Smith can be contacted at jls2@purdue.edu or 49-61279. For more information visit the program website.
“We are excited about our collaboration with Professor Sudhoff,” says John Campbell, the associate vice president who heads ITaP research computing. “Community clusters already are a proven cost-effective system for advancing research on campus. This is an example of how they contribute to campus efforts to reduce energy consumption as well.”
ITaP still has limited capacity available in the new Hansen community cluster, which went online in September. The new Carter community cluster, the fastest campus supercomputer outside of a national lab, will be available to Purdue researchers in the spring. Faculty members interested in the clusters can contact rcac-cluster-purchases@lists.purdue.edu.
Sudhoff moved his computation to ten nodes on the Rossmann cluster, giving him 240 processing cores. His retired system required 108 nodes to provide 216 cores, which not only weren’t as energy efficient as Rossmann’s but also were less powerful computationally.
The Sudhoff lab is using Rossmann to run genetic algorithms aimed at optimizing component and system performance, for example in the design of electric motors and generators. The research also applies to power system control — particularly for smart grid applications like shipboard systems, one of Sudhoff’s specialties — and to vulnerability analysis of shipboard integrated engineering plants, among other things.
Writer: Greg Kline, science and technology writer, Information Technology at Purdue, 765-494-8167 (office), 765-426-8545 (cell), gkline@purdue.edu