| |
Home
> Archives > Additional
Archives > Jim Lindelien |
| |
Jim Lindelien
TimeLogic Corporation
Crystal Bay, NV 92093
jiml@timelogic.com
http://www.timelogic.com
Abstract:
Server farm speed that fits on your desktop? Reconfigurable Computing
yields Substantial Performance and Price-Performance Improvements
for Biosequence Analysis
A conventional server augmented with accelerator cards employing
Field Programmable Gate Array (FPGA) technology yields substantial
gains in absolute speed and price-performance for large-scale biosequence
analysis requiring Heuristic or Dynamic Programming algorithms. Popular
examples of these algorithms include BLAST, Hidden Markov Models,
Profilesearch, Smith-Waterman, and Framesearch. Rapid on-chip reconfigurability
of FPGA device technology (so called "soft hardware") allows
each analysis to run on a parallel array of ideal algorithm-specific
processors and data path topology. The processors themselves and the
topology interconnecting them is synthesized in real-time just prior
to the start of each task, and persist only for the duration of each
analysis. Software running on the host server's CPUs controls overall
job flow, FPGA chip logical redefinition, and pipelines pre- and post-process
CPU loads during accelerated bulk query-vs-database combinatorics
running on the FPGAs. Data storage and transport bottlenecks commonly
encountered in academic bioinformatics software are avoided by design,
permitting very fast accelerated speed on a single small server, without
costly RAM caching of entire large genetic databases. Wall clock relative
speed gains of two to three orders of magnitude per server are realized,
allowing one such server to replace or reclaim computing capacity
equal to a large scale farm. |