Confusion about process and thread settings on the cluster about smilei HOT 2 CLOSED

Hi

First of all you can read the closed issue #36 which is probably the same kind of problem you encounter. Most problems reported to us similar to yours are simply due to a bad openMP threads positioning. If you don't know what I am talking about, refer to your system administrator.

I strongly encourage not tu use openMP across sockets. Your cluster probably has 2 sockets per node so I would use 2 MPI per nodes instead of one and 8 openMP threads per MPI rank instead of 16.

There is a typo in your submision file. You probably want to execute module purge and not "module puege`.

The number after -n is the number of MPI proceses you want mpirun to spawn. In your case that should be twice the number of nodes you are using. The number of threads is then set with the environment variables OMP_NUM_THREADS that you have to set to 8 in your case. If you don't set it, the behaviour depends on the system.

Again, I can not emphasize enough that you have to be very careful about your thread positioning. Bind them to core, not more than 1 per core and all 8 threads on the same socket.

The error you are reporting at the end of your issue seems to be related to a bad MPI installation. Have you ever run any MPI program on this cluster ?

Last piece of info: you can completely disregard the "Load" for now. It is just a virtual quantity we use for dynamic load balancing. It is completely independent to your problems.

Hope this helps solving your problems and see you back here with questions/issues about the code itself.

Cheers

from smilei.

After reading your reply and yesterday's attempt, I understand how to set these variables now. The typo is just in the dialog box above. I do want to execute module purge. Thanks for correcting this mistake. I have run MPI program but without OpenMP. So, to make smilei run on our cluster, I have to do some extra settings.
Thank you for your teaching! I will come back soon.

from smilei.