[Oscar-users] Oscar 4, torque and mpiexec

[Salvatore Di Nardo]

Hi all, i'm having a problem running mpi jobs with torque.

Problem is that if i submit this script:

===========================================================
#!/bin/sh
#PBS -S /bin/sh
#PBS -m e
#PBS -l cput=500:00:00
#PBS -l nodes=3:ppn=2
cd /home/salvator
time mpiexec -boot -v -n 6 mpiblast -p blastx -d nr -i frag.0 -o frag.0.out6p_pbs
===========================================================

All 6 processes will be processed by a single node ( each node have 2 procs) and are not splitted on 3 nodes. After some feedback from Torque mailing list seems that answer is that exists several version of mpiblast and mpiexec, and not all can retrive automatically nodelist from pbs, but if i understood well, the version that comes with Oscar 4 should have the right version. If is this the case why not works? And if i need to recompile the correct version of mpiexec, in README file i read that i must also patch torque sources, but as far i know the version of torque that comes with oscar is modified and dunno it i should patch it or not.

What i must do, to let work torque correctly? Please, it's realy important for me to be able to run mpi jobs avoiding to start manually lamboot and giving nodelist statically ( giving a static list, will force to use those nodes, also if was alseady rederved by torque to other jobs, or will not mark it as already used).

Salvatore Di Nardo