The Torque Batch System
There are currently 128 job slots allocated by the Torque batch queue system. Use of Torque (previously PBS Pro) is very similar to LSF. Users wishing to use Torque will need to alter their environment to give priority to the Torque binaries:
# bash users put this into ~/.bashrc
PATH=/opt/torque/bin:/opt/torque/sbin:${PATH}
MANPATH=/opt/torque/man:${MANPATH}
# tcsh users put this into ~/.cshrc
setenv PATH /opt/torque/bin:/opt/torque/sbin:${PATH}
setenv MANPATH /opt/torque/man:${MANPATH}
To submit a serial job (single process) just use the qsub command: "qsub ./job.sh"
To check a submitted job you use the "qstat" command.
For parallel jobs it is easiest to use a job submit file like the following:
=======================
#PBS -l nodes=32:ppn=4
cd /home/gragghia/mp_linpack_10/bin/russ
mpiexec -comm mpich-ib ./xhpl
=======================
Here I have requested that the job run on 32 nodes each with 4 processes
(a total of 128 processes). I first change to the working directory of
my program, and then I use the "mpiexec" script to execute the mpi
program. I have indicated to mpiexec that my program uses the
infiniband network for MPI. The ./xhpl executable was originally
compiled with MPICH and was used previously with the LSF batch system.
I was able to use it with PBS without recompiling, so I don't expect that
any programs will require modifications if they are submitted in this way.
|
|
 Copyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding USG HPC? Send feedback
|
