LIVEwire News Briefs:

Linux Supercomputer Delivers 11.1 GB/s Parallel I/O

Tungsten, the top-ranking Linux cluster recently installed at the National Center for Supercomputing Applications (NCSA) at the University of Illinois, has achieved 11.1 Gigabytes per second of I/O throughput. This complements the cluster's 9.8 teraflops of computational capability, which was recently reported on the November 2003 Top500 Supercomputer Sites list. This performance level was reached using the Lustre file system for simultaneous writes to individual files from 104 clients across 120 terabytes of DataDirect storage.

"High-end applications in science and engineering are increasingly data intensive, and are requiring significantly more I/O capability than we have had available in the past. As clusters grow and applications become more data-oriented, the I/O capability is a much more significant component of the system," said Rob Pennington, head of NCSA's Computing and Data Management Directorate. "We expect the new cluster with the Lustre file system to fulfill a real need for the applications, which is good news for the scientists."

"NCSA is known for its leadership in high-performance computing, and we are very pleased to support their effort to develop a balanced general purpose supercomputing architecture," said Phil Schwan, chief executive officer of Cluster File Systems Inc. "Lustre, with its scalable object-based architecture and intelligent protocols and serialization, is ideally suited to the task of delivering dramatic increases in I/O throughput to meet the needs of our customers."

NCSA's Tungsten cluster was designed to provide the large-scale computational and I/O capabilities that are necessary to continue the advance of breakthrough scientific research. Tungsten, with more than 1,450 dual-processor Dell PowerEdge 1750 servers running Red Hat Linux, employs the Lustre file system across a 104 node I/O sub-cluster with more than 120 terabytes of DataDirect storage. Tungsten is ranked fourth on the latest Top 500 supercomputers list and is the most powerful Linux cluster on the list.

"DataDirect Networks is very excited to have delivered and installed tens of teraflops and hundreds of terabytes of SwiftCluster-based storage networking to our valued HPC customers since the beginning of the year", said Alex Bouzari, CEO of DataDirect Networks. "DataDirect's SwiftCluster architecture, optimized for the Lustre file system, achieves its goal of delivering highly scalable data processing with very low latency."

The Tungsten cluster will provide a boost in productivity for the researchers who use NCSA computing facilities to investigate many of science's most important questions -- the large-scale structure of the universe, the properties of stars, the flow of fluids through channels, the dynamics of biological systems, the design of engineering structures and the nature of matter itself.