Parallel processing in bioinformatics and molecular modelling
Gordon D. Pusch
gdpusch at NO.xnet.SPAM.com
Fri Apr 26 19:10:16 EST 2002
howardolson at ausi.com (Howard Olson) writes:
> Are there adequate parallel software systems for bioinformatics work
> in any Cray system?
Crays are ``vector'' machines with a relatively long pipeline, which
means that they do not perform at all well unless you are performing
=EXACTLY= the same sequence of calculations on every member of a
large set of numbers, and none of those calculations are conditional
on any intremediate result. By contrast, nearly everything in bioinformatics
is ``trivially parallelizable'' rather than vectorizable, but involves
a lot of conditional branches, e.g., during the backtracking phase of
a sequence alignment, when you need to choose the path with the best score
through the alignment matrix. Hence, Crays are the exact =OPPOSITE=
of what you want for most bioinformatics applications; what you really need
is a ``Beowulf'' style massively parallel cluster of independent processors.
> If not, is there any supercomputer system available commercially with
> such software available.
Timelogic.com <http://Timelogic.com/> sells a special-purpose computer
that has a built-in FPGA-based hardware accelerator. By extracting the
time-critical ``inner loop'' of the bioinformatic algorithm and
implementing it in field-programmable firmware rather than software,
they can achieve more than a two orders of magnitude speedup.
> I am particularly interested in the quantum biochemistry of the CNS.
I cannot help you there.
-- Gordon D. Pusch
perl -e '$_ = "gdpusch\@NO.xnet.SPAM.com\n"; s/NO\.//; s/SPAM\.//; print;'
More information about the Comp-bio