This week the San Diego Supercomputer Center announced that it is in the evaluation phase for a mid-range supercomputing resource that employs flash memory for large scale computing. The system is actually part of an existing supercomputer at the center
The new High-Performance Computing (HPC) system, dubbed “Dash,” is an element of the Triton Resource, an integrated, data-intensive resource primarily designed to support UC San Diego and UC researchers that went online earlier this summer. As envisioned, this “system within a system” will help researchers looking for solutions to particularly data-intensive problems that arise in astrophysics, genomics and many other domains of science.
Dash is a 5.2 TF partition of Triton with 68 Appro GreenBlade servers armed with dual-socket quad-core Nehalems and an IB interconnect
…it has several unique properties, including the first use of flash memory technology in an HPC system, using Intel High-Performance SATA Solid-State Drives. Four of its nodes are specially configured as I/O nodes each serving up 1 terabyte (TB) of flash memory to any other node, courtesy of new I/O controllers also developed by Intel Corporation and integrated by Appro International, Inc.
Dash’s flash will be used for fast file access and swap — which makes flash look like fast disk. Interestingly they are also using ScaleMP’s vSMP software
In its current configuration, Dash has 48 gigabytes (GB) of DRAM memory on each node, and employs vSMP Foundation software from ScaleMP, Inc. that provides virtual symmetric multiprocessing capabilities and aggregates memory across 16 nodes into shared memory “supernodes,” giving users access to as much as 768 GB of shared DRAM memory in addition to 1 TB of flash memory per “supernode”.
…”Dash can do random data accesses one order-of-magnitude faster than other machines,” said Allan Snavely, associate director at SDSC. “This means it can solve data-mining problems that are looking for the proverbial ‘needle in the haystack’ more than 10 times faster than could be done on even much larger supercomputers that still rely on older ‘spinning disk’ technology.”
See HPCwire for the full release, in which you will find more information.