I am using a Linux HPC server and am kind of the manager.
Recently one of the users ran a multithreaded job using OpenMP. All jobs are supposed to be submitted through the scheduler (SunGridEngine, SGE). One user submited a multi-process (MP) job and another user submited a multi-threaded (MTH) job.
For the MP job, CPUs are assigned at the scheduling stage so there is no problem at all. But I wonder how CPUs are assigned for the MTH job. And I am worried if a CPU occupied by the MTH job will interrupt the CPU used by the MP job.
Will it be alright? If not, how can I manage this situation?
FYI: Sometimes, the load exceeds the number of CPU cores when using MTH. So I tried to set the number of threads (limit) at the level of OMP coding. But the load (CPU %) frequently exceeds CPU load a little bit.