We learned a pretty hard lesson about the after_each_exec step in the
MPI Details section over the past few weeks, particularly when used
in conjunction with SLURM's "srun" command (although this is not a
SLURM-specific issue).
I highly encourage everyone to read the commit message for r657 -- it
changed what we do at Cisco in the after_each_exec step:
https://svn.open-mpi.org/trac/mtt/changeset/657
--
Jeff Squyres
Cisco Systems
|