Matt,

I just found something else with your job script. You are defining the
LD_LIBRARY_PATH environment variable, and *then* loading the openmpi
environment module:

>>>         #!/bin/bash
>>>         export TMPDIR=$SCRATCH/abyss_tmp/
>>>         LD_LIBRARY_PATH=/work/01301/mmacmane
>>>         LD_LIBRARY_PATH=/work/01301/mmacmane/bin
>>>         LD_LIBRARY_PATH=/work/01301/mmacmane/include
>>>         LD_LIBRARY_PATH=/work/01301/mmacmane/etc
>>>         LD_LIBRARY_PATH=/work/01301/mmacmane/lib
>>>         LD_LIBRARY_PATH=/work/01301/mmacmane/openmpi-1.4.1
>>>         cd /work/01301/mmacmane/Ray-0.0.6
>>>         module load openmpi

This is bad. Not only are you incorrectly setting your LD_LIBRARY_PATH,
but after you modify it, the command

module load openmpi

is overwriting your customizations with what I presume is the default
evironment for the officially supported openmpi version at TACC/on Ranger.

Also from this line:

>>>             Module mvapich superceded

it looks like you already had the mvapich module loaded before you
submitted your openmpi job with qsub. This isn't a good idea, either. I
don't know how your modules are written, but it's possible that your
mvapich module has settings that are incompatible with your home-grown
openmpi setup.

You should make sure you remove your mvapich module before submitting
your script. I don't use GNU environment modules (yet), so you'll have
to contact your local system admin or read the module man page to learn
how to remove the mvapich module.

Just to be clear, these problems are problems with your basic Linux/Unix
environment, and aren't really SGE- or OpenMPI-specific problems.

-- 
Prentice

Reply via email to