Lustre has multiple controllers, which means that there is less chance of having a bottleneck when lots of sources are seeking or writing data to disks. It has been shown to support more than 50,000 clients simultaneously. Lustre can deliver over 2GB/s (gigabits per second) to an individual client. It is an open-source technology, though in 2007 Sun Microsystems acquired CFS, the company that did much of the early development work.
"Lustre is a good, scalable, robust file system," said Mark Seager, principal investigator for supercomputing platforms at the U.S. Energy Department's Lawrence Livermore National Laboratory (LLNL). About 25 of the lab's systems, both for classified and unclassified work, run on the file system.
LLNL was the first large organization to use the file system in a production development, and continues to fund its development. LLNL also turned out to be one of Whamcloud's first customers. Last week, LLNL commissioned Whamcloud to undertake a number of Lustre-related tasks. One is upgrading the LLNL's LMT (Lustre Monitoring Tool) so that it can work with Lustre 2.0. A debugging tool, LMT allows an administrator to determine if one particular application is taking up too many resources.
"In a multi-user environment, when a file system gets too busy, it is hard to correlate what application is causing that," Gorda said. LLNL will make this software open source, and it is scheduled to be released early in 2011.
Whamcloud will also help LLNL characterize Lustre performance on solid-state disks, which are expected to be widely used in HPC systems as their storage needs continue to balloon. "We expect that we'll find performance issues inside Lustre that would be addressed in the long term," Gorda said. "We're getting a metric for determining if Flash [drives] are the right thing to use."
In contrast to Gorda's claims, LLNL's Seager has not felt that Lustre support has lagged since Oracle assumed control of the technology. Oracle has helped LLNL in matters of hardware maintenance and improving the way the failover mechanisms work. "We're very happy with the collaboration with Oracle," he said.
Still, Whamcloud plans to ramp up support for the technology.
"There's been this confusion for the past six months, and I think we're seeing a calm return to the community. People are seeing that Lustre is not going away," Gorda said.