More details about the hardware held at the Los Alamos Nuclear Laboratory has emerged: it uses Panasas clustered file storage for its petascale nuclear weapons testing supercomputer.
A petascale supercomputer computes at the petaflop level - ten to the power 15 or one million billion floating point operations per second. Parallel supercomputers need parallelised I/O so that the many computing elements can be fed data fast enough to keep them operating at maximum speed.
Serial network-attached storage (NAS) products are not fast enough and clustered file system products, such as Isilon's have emerged to speed data delivery
LANL set up the very first (Cray) supercomputer which ran at 80Mflops. It now uses Panasas storage to supply seven terascale - meaning the one thousand billion FLOPS level - supercomputers with data, and is building the Roadrunner hybrid petascale supercomputer from AMD Opteron and IBM Cell processors, both running Linux.
A single Panasas ActiveStor storage cluster is shared globally between LANL's terascale supercomputers. There is more than 1PB of storage capacity with approximately 50Gbit/s of input/output. "We've been using Panasas storage for a long time at LANL to provide scalable and globally-shared storage to multiple terascale clusters," said Gary Grider, group leader for the High Performance Computing (HPC) systems integration at Los Alamos. "We will leverage (it) to provide the I/O solution for the Roadrunner system."
Interest in petascale computing environments is growing with a Petascale Data Storage computing institute being set up. There is a trend for petascale HPC to move to extreme enterprise computing environments such as Google, Ask.com, MS Live, Amazon, Weta Digital, and so on. As both data size and data rates need to grow for more general business applications then parallelised and clustered I/O will be seen there too.
Sun, for example, is offering clustered storage and embedded processing to the medical imaging market with its StorageTek 5800 product.