The clock is running on the goal to stage an exascale technology demonstration capable of handling a peak of 400 petaflops by 2015. This is more data than a single stream can deliver, so the search is on in the HPC community for a solution to these I/O demands. The hardware approach, writes Brent Gorda, Whamcloud CEO and President, in his HPCWire article Why Lustre Is Set to Excel in Exascale, -- boosting performance on the single unit and going widely parallel -- will work until power limitations and parallelism issues become an impediment. Gorda enthusiastically nominates the Lustre file system to fill the software side of the equation. Lustre, he continues, based on an object store, has the necessary fundamental architecture for exascale, which the HPC file system technologists think will be based on an object store approach. It might work, Gorda continues, to introduce new file types to Lustre that will provide exascale object storage semantics internally. While this will require development of the underlying object model, Gorda points out that it holds the promise that the same file system will be able to support the full range of applications from (legacy) POSIX through to exascale. Finally, under development since 1999, Lustre has the maturity and stability that developers seek as the starting point for rapid and diverse development, Gorda concludes.
Lustre Strategy Update
Read More ...