SCIENCE

Whamcloud and LLNL will work together on the Lustre Monitoring Tool (LMT), a debugging tool for administrators in supercomputing environments, as well as on Lustre performance testing utilizing Solid State Devices (SSDs) on LLNL's large test cluster Hyperion Data Intensive Testbed. That testbed has been measured at 46M IOPS @ 4K blocks and over 500GB/s for 1MB blocks with local files systems. The impact of SSDs on performance on a wide scale is broadly anticipated to be significant. These at-scale tests will be a first with the Lustre global parallel file system.

"We've been long-time supporters of Lustre and are extremely happy to see that Whamcloud is continuing to support the Lustre on Linux community," said Mark Seager, Livermore's principal investigator for ASCI platforms. "LLNL continues to be at the forefront of developing HPC technologies and pushing the envelope for data intensive computing environments that support our national security mission."

"We're excited to be working with Livermore, a preeminent supercomputing site, as this will directly benefit the Lustre community. It's an opportunity to extend Lustre functionality for HPC administrators with an incredibly knowledgeable and talented partner," said Brent Gorda, CEO of Whamcloud. "No other at-scale tests have been performed like this, and the results will be made available for everyone to use."

The LMT watches system hardware and any current processes and presents it to the administrator in an easily digestible format. This is especially useful in debugging. Whamcloud will work with LLNL to extend the LMT to Lustre 2.0, the most recent version of Lustre. These improvements will be included in the Whamcloud testing and build process and will be made available worldwide.

LLNL's large test cluster Hyperion has recently added a large amount of PCI-based SSDs for storage to create a data intensive testbed. SSDs are seen as the future of storage in HPC. In this challenging test environment with minimal latency and high bandwidth rates between devices, Whamcloud and LLNL will run a wide range of performance tests and will incorporate improvements directly into Lustre 2.0. This will help extend Lustre onto large petaflop systems like the upcoming 20 petaFLOP/s Sequoia system at LLNL, a third generation BlueGene system, which will require huge data bandwidths.