Code_Saturne on hpc

[Julien.Commanay]

Dear support team,

I'm currently trying to transfer runs prepared locally (using n_threads=xxx in run.cfg)) on a machine to hpc. When using slurm and restarting the calculation on hpc, I'm using n_procs= yyy in run.cfg but I'm facing errors due to the fact that the folder already exists in RESU. Since I deleted it before starting the run, I have the feeling each processor is trying to create the folder what leads to the error. Any advice on this point?

Thanks in advance for your support

JCo

[Julien.Commanay]

Hello,

I have made an additional test locally on my machine. I'm asking 5 processes with 1 thread per process and when looking at my log file, I can read:
Processors/node: 6
OpenMP threads : 4
I thought I would have respectively 5 and 1.
Could someone explain me this?
Kind regards
JCo

[Julien.Commanay]

And when looking at my run.cfg file, I have n_procs =5, what makes sense regarding what I initially wanted.

[Yvan Fournier]

Hello,

Could you follow the forum usage recommendations and provide the relevant logs ?

On the cluster, are you starting the code from the GUI or "code_saturne submit" ? Is the post-install (etc/code_saturne.cfg) step done ?

Regards,

Yvan

[Julien.Commanay]

Hello Yvan,

Please accept my apologizes if I have not followed the forum usage recommandations and provide the relevant logs.

I've performed several tests localy with and without the GUI to understand some points and performance issues (much clearer now for procs and threads) and without the GUI on the cluster. I have contacted the team that performed the installation of Code_Saturne 8.03 on the cluster (installation done a few days ago) to check some points related to their installation.

I come back to you (with logs and relevant information) if this issue persists.

Thanks again

Julien

[Julien.Commanay]

Hello,

This message to confirm it was an installation issue on the cluster, everything seems to work fine now.

Kind regards