When such a test is done, please note that in my view it is possible to achieve a setup when MPI hello world job runs on a non-shared filesystem. But imagine an MPI job that writes to some output files from each WN, and then expects that all those files are present at the same directory. On a non-shared filesystem such a job will fail. Therefore a monitoring job should check for such a situation. Also note that the fact that a job involving 4 CPUs runs doesn't mean that a job involving all the published CPUs will be allowed to run, and succeed. Emanouil Atanassov [log in to unmask]