[Bioclusters] mpiblast on SGI Altix

Aaron Darling darling at cs.wisc.edu
Wed Mar 16 12:48:41 EST 2005

Since XFS seems to provide these machines with extremely fast shared 
storage I would suggest setting your "Local" directory equal to the 
"Shared" storage directory.  Then you can avoid all the fragment copy 

Also, a google for mpiblast sgi altix 
(http://www.google.com/search?q=mpiblast+sgi+altix) reveals what appears 
to be some excellent work done by folks at ORNL and NCSU to optimize 
mpiblast for the altix architecture.  It may be worthwhile to contact 
them for more  info...
They benchmarked version 1.2.1.  It would be interesting to see how 
their mpiblast benchmark results improve on an altix with shared and 
local set identical and with the --disable-mpi-db flag.


Joe Landman wrote:

>Hi Mike:
>  Make sure you configure mpiblast to use the shared memory device (ch_shmem)
>rather than the ethernet (ch_p4), or similar devices if you are using MPICH. 
>Dont worry about assignments of partitions to nodes, that should not be an issue
>for this machine.
>Joseph Landman, Ph.D
>Scalable Informatics LLC,
>email: landman at scalableinformatics.com
>web  : http://scalableinformatics.com
>phone: +1 734 612 4615
>---------- Original Message -----------
>From: Mike Muratet <muratem at eng.uah.edu>
>To: mpiblast-users at sourceforge.net, bioclusters at bioinformatics.org
>Sent: Wed, 16 Mar 2005 11:13:28 -0600 (CST)
>Subject: [Bioclusters] mpiblast on SGI Altix
>>Does anyone have any experience with mpiblast on the SGI Altix and can 
>>you offer any suggestions for configuring jobs? The Altix 350 has 16-
>>CPU nodes with shared memory. The machine I'm using has 56 total CPUs. 
>>Can anyone offer any advice for how many database partitions to use, 
>>and how they get assigned to nodes/CPUs?
>>Bioclusters maillist  -  Bioclusters at bioinformatics.org
>------- End of Original Message -------
>Bioclusters maillist  -  Bioclusters at bioinformatics.org

More information about the Bioclusters mailing list