There are too many diverse scenarii about the communication and execution of
parallel applications: the code can be a MPI code or composed of different
interacting programs possibly launched via
ssh on every nodes; input and
output files can use NFS if this file system is present, or they can be
splitted and uploaded to each node participating to the calculus.
Then, we will see: what supplementary information has to be provided in the server configuration file; how to write a batch submission meta-script in a SeD; and how to record the parallel/batch service.