benchmarks or experience with blast/fasta on different cpus? especially cache impact.

Haruna Cofer haruna at columbus.sgi.com
Fri Oct 29 17:46:09 EST 1999


Hi Joel,

At SGI (sorry, not SUN!), we have evaluated the performance of both
BLAST and FASTA on our various MIPS processors with cache sizes ranging
from 1MB to 8MB.  The result is that there is usually very little (less
than 5%) performance difference between the 1MB vs 8MB cache sizes. 
Both BLAST and FASTA maintain good locality of reference in their codes,
so in general, the cache size is not so much an issue as is the speed of
the processor.  However, if your query sequences are relatively long (as
in thousands of letters), then you may see some cache effects with the
smaller cache sizes.

If you are interested, we have a Bioinformatics Performance Report that
contains this type of comparative information (but only for SGI
systems!), along with information about our High-Throughput BLAST and
Parallel CLUSTALW (if you are interested in performance enhanced
versions of those):

 http://www.sgi.com/chembio/bioperf99.pdf

This report also contains BLAST and FASTA benchmarks for our new LINUX
server called the SGI 1400, which is a 4-CPU 500MHz 2MB cache Pentium
III Xeon system. 

And finally, we also have technical information (including system
requirements and porting notes) about running BLAST and FASTA (and
CLUSTALW) on SGI systems here:

 http://www.sgi.com/chembio/resources/blast/
 http://www.sgi.com/chembio/resources/fasta/
 http://www.sgi.com/chembio/resources/clustalw/

Thanks, Joel, and please do let me know if you have any questions!

-- Haruna :)

-- 
-------------------------------------------------------------------- 
   _editted_at_bionet_
  _editted_at_bionet_
  _editted_at_bionet_
  _editted_at_bionet_
  _editted_at_bionet_
--------------------------------------------------------------------




More information about the Bio-soft mailing list