SGE script to start a parallel job on the cluster

To run a parallel MPI program on the cluster :

• Create a dedicated directory somewhere in /work
• Place all files needed by your job (and possibly also the executable) in this directory.
• Create a file with a suitable name (like myjob.sh) containing the script that follows.

#!/bin/csh -f # # # The name of the job (MPI_job) ... # #$ -N MPI_job # # The parallel environment (mpi_fast) and number of processors (9) ... # #$ -pe mpi_fast 9 # # For the 'mpi_fast' environment you may want to define a master queue : # #$ -masterq server.q # # The version of MPICH to use, transport protocol & a trick to delete cleanly # running MPICH jobs ... # #$ -v MPIR_HOME=/usr/local/mpich-ssh #$ -v P4_RSHCOMMAND=rsh #$ -v MPICH_PROCESS_GROUP=no # # Execute from the current working directory ... # #$ -cwd # # Standard error and output should go into the current working directory ... # #$ -e ./ #$ -o ./ echo "Got $NSLOTS slots." $MPIR_HOME/bin/mpirun -np $NSLOTS -machinefile $TMPDIR/machines my_MPI_program

• Edit this file and change the obvious things (like program name, arguments to program, job name, number of processors).
• Submit the job with qsub myjob.sh.
• Check that it started without problems using qstat.