[Beowulf] CCL:dual-core Opteron

David Mathog mathog at mendel.bio.caltech.edu
Thu Jul 7 11:23:19 PDT 2005


>   I am sure it is possible to find codes and test cases which will
fill the
> bandwidth for the memory controller.  The usual suspects (Amber8, GAMESS,
> BLAST, HMMer) did not.

Regarding HMMer, did you try running tests where you step through
a range of query sequence sizes (all vs. the same HMM database)?
On the Athlon MP 2200+ machines in our cluster there was
a large variation in run time resulting from complex cache interactions.
This processor has a relatively small cache and is presumably much
more sensitive to this than would be the Opterons.  A substantial
amount of reorganization of the data structures minimized this effect
but at least initially going from n residues to n+1 residues in the
same sequence could change the run time for the query by a factor of 3.

You may be able to trigger the same effect on an Opteron by using
a largish query sequence, say 4000 amino acids or so.  Admittedly
there are not very many proteins which are that large.


Regards,

David Mathog
mathog at caltech.edu
Manager, Sequence Analysis Facility, Biology Division, Caltech



More information about the Beowulf mailing list