SCIENCE

A paper co-authored by Dr. Andrew Lumsdaine, director of the Open Systems Lab (OSL) at the Indiana University Pervasive Technology Institute, and former OSL researchers Torsten Hoefler (University of Illinois at Urbana-Champaign) and Timo Schneider (University of Technology Chemnitz) has received Supercomputing 2010's Best Paper accolade. As the premier annual international supercomputing conference, Supercomputing 2010 drew thousands of participants from all over the world.

The paper, "Characterizing the Influence of System Noise on Large-Scale Applications by Simulation," analyzes the impact of system noise, or communication delays, on the performance of large-scale applications running on multiple computer processors connected over a network.

"Torsten did an excellent job presenting this paper," said Lumsdaine. "We were up against papers from some of the best and brightest minds in supercomputing. We are honored to receive the top award."

The researchers found that system noise impacts applications running across slower networks less than those running across faster ones. When system noise creates a bottleneck, it eliminates any advantage a faster network may have. This finding is crucial for the design of large-scale systems, because the noise bottleneck must be considered when a system is designed.

"The committee believes that this is likely to be the definitive paper in a debate on the impact of system noise that has been going on for a decade," added Barbara Chapman, technical papers co-chair at Supercomputing 2010.

Additionally, the team proposed a model that suggests that non-blocking operations can be used to reduce noise. Non-blocking operations are those that can proceed even if a previous operation is suspended, thereby eliminating the need to wait for a previous action to finish.

The full paper is available at http://portal.acm.org/citation.cfm?id=1884668