DATE: Thursday, October 7, 2004
TIME: Noon - 1 pm
PLACE: Wean Hall 8220

Peter Braam
Cluster File Systems, Inc.

The Lustre File System

Lustre started as an academic research project and is now powering many super computers. A single, petabyte scale Lustre file system will be deployed at LLNL early 2005 to provide the BlueGene super computer and some 10 other clusters with scalable file storage. Another deployment targets the 8,000 client Red Storm cluster. Both deployments target 50GB/sec of aggregate I/O throughput. This presentation will review some of Lustre's design, particularly in the area of dealing with very large file systems and high I/O efficiency. We will then highlight some of the process and pain we have gone through to reach success on major deployments.

Peter Braam is founder and president of Cluster File Systems, Inc., which has developed the Lustre file system that now powers many of the worlds largest super computers and has achieved outstanding bandwidth and scalability goals. Peter is a specialist in distributed file systems, best known for his work on the Lustre project and on the Coda and InterMezzo file systems, which won awards at LinuxWorld and the Open Source Convention in 1999. He received his PhD in 1987 and subsequently held tenured faculty positions at Utah and Oxford where he began to teach Computer Science in 1992. He joined Carnegie Mellon's faculty in 1996 where he led the Coda project for 3 years. In 1999 he began to commit most of his time to the Linux industry as a part time Cluster and File Systems Architect for Red Hat. His contracts with the Storage industry and National Laboratories focused on solutions to advanced problems, with an emphasis on Linux.

Visitor Host:
M. Satyanarayanan

Visitor Coordinator:
For appointments, contact Tracy Farbacher

For Further Seminar Info Contact:
or visit