This is a short overview of Lustre, a new open source cluster file system. The name Lustre embodies "Linux" and "Cluster." Lustre focusses scalability for use on large compute clusters, but can equally well serve smaller commercial environments. Lustre runs over different networks, including at present Ethernet and Quadrics.
Lustre originated from research done in the Coda project at Carnegie Mellon. It has seen interest from several companies in the storage industry that contributed to the design and funded some implementation. Soon after the original ideas came out, the USA National Laboratories and the DOD started to explore Lustre as a potential next generation file system. During this stage of the project we received a lot of help and insight from Los Alamos and Sandia National Laboratories, most significantly from Lee Ward.
Lustre provides many new features and embodies significant complexity. In order to reach a usable intermediate target soon, Mark Seager from Lawrence Livermore pushed forward with Lustre Lite. We are hopeful that Lustre Lite will be the shared file system on the new 800 node MCR Linux cluster during 2002.
This paper provides a high level overview of Lustre.