Anyone: NCBI BLAST2 on >1 CPU on Solaris?

Peter C. Tribble ptribble at hgmp.mrc.ac.uk
Thu May 4 02:54:18 EST 2000


In article <8er2lf$um8$1 at nnrp1.deja.com>,
	mike4 at cris.com writes:
> I'm having trouble getting BLAST2 (executable: blastall) to run on more
> than one processor (i.e., when I specify the number of processors as 4
> with the '-a' command line option).  It could have something to do with
> my local environment, although I can get other programs (e.g. Perl) to
> utilize all 4 processors.  Has anyone observed BLAST-2 running on more
> than one processor in a 64-bit Solaris environment?  Does anyone know
> of a trick to get BLAST2 to compile so that it can use all the
> processors?

We normally seem to use -a 8, as far as I can tell watching the queues,
and this normally means that 8 threads run - so on a 20-way box, we
actually see 8 processors used.

There are times in blast when it's essentially single threaded (I think
when it's actually finished the search and is extracting the results is
one), but the search phase uses multiple processors very nicely.

If you've got top, it will show the number of threads. This isn't equal
to the number of threads you've told it to use (in our case, with -a 8
we see 13 threads, which drops to 5 at the end of the run). So you can
tell from that whether it's actually creating the threads or not (and
what stage of the processing it's at).

If you're I/O bound then you'll never see high cpu usage, even with
multiple threads, as they're all waiting for the data. So it may then
appear that it's effectively only using a single cpu.

-- 
-Peter Tribble
HGMP Computing Services
http://www.hgmp.mrc.ac.uk/~ptribble/





More information about the Bio-soft mailing list