From Engineering Grid Wiki
OpenMPI is located at /cluster/cloud/openmpi.
After compiling your MPI program with the compilers in that directory (or having us install it for you) you need to submit the job through the correct Parallel Environment to be picked up as an MPI job.
In this case, that PE is 'openmpi', and a job script utilizing it will look somewhat like this:
#!/bin/sh #$ -q all.q #$ -N openmpi_jobname #$ -pe openmpi 10 #$ -cwd #$ -v LD_LIBRARY_PATH=/cluster/cloud/openmpi/lib #$ -v HYDRA_ENV=all /cluster/cloud/openmpi/bin/mpiexec /absolute/path/to/your/executable
The number '10' in the line
#$ -pe openmpi 10
is the number of CPUs you want this job to use.
The openmpi parallel environment is set to 'fill up' nodes, so that as many jobs start on the same node as possible. Note the Cloud Cluster currently has no fast interconnect, and your parallel jobs will be running over Gigabit Ethernet when they have to operate over multiple nodes.