Printer friendly version

Los Alamos supercomputing cluster software wins prize

Researchers in the Laboratory's Advanced Computing Laboratory have been honored for their advances in connecting thousands of computers together to create clusters with much of the power of traditional supercomputers at a fraction of the cost.

Los Alamos' Clustermatic 4 software suite won top honors for Open Source Cluster Solution in the recent ClusterWorld Conference and Expo in San Jose, Calif. The winners were credited with developing technologies that have reshaped the information technology landscape and brought supercomputing power to a variety of new users.

"This is a wonderful and much-deserved recognition of the innovative work done by the Cluster Research team," said Bill Feiereisen, Computer and Computational Science (CCS) Division leader. "The Clustermatic 4 software includes a range of novel techniques that have for the first time been combined into a highly scalable clustering system. Clustermatic 4 is used widely by government, industry and academia, and it already is having broad impact on cluster computing."

The Advanced Computing Laboratory's Clustermatic project began four years ago to make it easier for scientists to link together off-the-shelf computer systems to create low-cost clusters that could plow through data at speeds approaching those of traditional supercomputers. The most recent version, Clustermatic 4, is an open-source collection of technologies that incorporate significant advances in scaling clusters to systems containing 2,000 or more processors. It was released in November 2003 and already is finding use in Los Alamos' Lightning supercomputer, an 11-trillion-operations-per-second Linux cluster.

Setting up clusters is extremely tedious and error-prone due to the inherent autonomy of the individual nodes and the huge scale of modern clusters. Los Alamos' Cluster Research team redesigned cluster system architecture starting from the low-level machine setup and extending to all aspects of system operation to increase reliability and efficiency and reduce autonomy. More information is available at http://public.lanl.gov/cluster/ online.

Judges for the competition included Douglas Eadline, editor in chief of ClusterWorld magazine and scientists and engineers from the National Science Foundation, Lockheed-Martin Corp., Indiana University and the Moscow Joint Supercomputing Center.

ClusterWorld magazine is dedicated to cluster-related topics; its Web site is http://www.clusterworld.com online.

Contributors to Clustermatic include Ron Minnich, Sung-Eun Choi, Erik Hendriks, Matt Sottile, Greg Watson and Li-Ta Lo, all of Los Alamos' Advanced Computing Laboratory.

Funding came from the Mathematical Information and Computer Sciences program of the Department of Energy's Office of Science, the Los Alamos Computer Science Institute of the Advanced Simulation and Computing program and the Laboratory Directed Research and Development program.

-- Jim Danneskiold

Other Headlines