"MPI-like" submission of jobs?

Hi all,

Most jobs on our cluster are large MPI jobs reserving anywhere from 96 to several hundred cpus. Submitting a large number of one cpu jobs as Frealign does makes it difficult for these users to schedule their jobs sometimes.

I was wondering if anyone had suggestions on how to submit a Frealign job which reserves a block of cpus on an sge cluster, and submits frealign jobs to those cpus without letting those cpus return to the pool of cpus until the job is finished (preferably without having to rewrite too many of the submission scripts).

Best,

Axel