[Bioclusters] FPGA in bioinformatics clusters (again?)

Kathleen kathleen at massivelyparallel.com
Mon Feb 13 11:33:45 EST 2006


On Sat, 2006-02-11 at 08:18 -0700, Kathleen wrote:
> There are alternatives to FPGAs that provide similar speedup and 
> scalability.

    I am very interested in hearing how to get a speedup of 700
    on a HMM search similar to what can be achieved with a single
    $30,000 FPGA card. I'm not sure what a 700 CPU cluster costs,
    but my feeling is that it will be somewhat larger than adding
    an FPGA card to an existing server.

(KE) It may be possible to get super-linear speedup.  Therefore, you could
probably use slightly fewer nodes and still get the performance of 700 CPUs.
Taking a software-based approach, while increasing your channel count per
node, can significantly increase performance and scaling to the point of
competing well with FPGAs at a fraction of the cost. Furthermore, your
end-to-end overhead is lower and the impact of failure is less when taking a
software-based approach. 

> If you can utilize your existing infrastructure, but increase 
> performance by several orders of magnititude ... with software ...

    I am also interested in how I can increase my performance 
    several orders of magnitude using my existing infrastructure.

    These sound like very impressive improvements.

(KE) Do you have a COTS-based infrastructure?  What software applications
are you using on this hardware? What apps create bottlenecks on your system?


- Alan Kilian <kilian(at)bobodyne.com>

Bioclusters maillist  -  Bioclusters at bioinformatics.org

More information about the Bioclusters mailing list