[Bioclusters] mpiBLAST memory leak?

[Iddo Friedberg]

I am running mpiBLAST 1.4.0 using mpich and SGE on a Rocks (4.2.1) cluster.
I have been getting what seems to be consistent memory leaks that crashed
nodes after about 5-6 hours of running. Memory consumption steadily
increases until the node from which the job is running crashes.

I tried setting P4_GLOBMEMSIZE to under sysctl.shmmax. Still no good.
/var/log or the SGE logfiles do not contain information. Any ideas?

Thanks,

Iddo

-- 

I. Friedberg

"The only problem with troubleshooting is that
sometimes trouble shoots back."