Unpredictable behavior of SM2021 - MPI

mf

Hello,

this week I faced some weird behavior of SM2021 (MPI-version). I haven't found out the reason until now.

In my study (I cannot attach .comm due to confidentiality) I use POURSUITE once. It seems that SM is not able to reload the glob-file when changing to the next stage. Then something like in the attached image appears (confidential stuff blurred, sorry): a huge window with a lot of text (once for each MPI process, I presume) that I cannot close. Also, the next stage does not show the turning hourglass, although something is still running.

Sometimes it seems, this behavior does not occur when I change the number of MPI-processes to, lets say 12 (16 in the image with error).

Does anyone else see this behavior and knows how to resolve? I know it would be easier to resolve if I attached .comm etc. In this case, I am not allowed to do that sorry.

Thank you,

Mario.

Attached file:
Bildschirmfoto_vom_2022-02-17_09-38-27_02.jpg, 1062625kb

rodrigofarias

mf Do you have a newer version of this MPI implementation for CA? Im currently testing this 2021 version.

jonasloenartz

mf wrote:

Does anyone else see this behavior and knows how to resolve? I know it would be easier to resolve if I attached .comm etc. In this case, I am not allowed to do that sorry.

Hello, I had the same issue today. I am using the container you provide on GitHub (so far no problems! thanks again for the work -- > https//github.com/emefff/Code-Aster-MPI-in-Singularity-of-SM2021). I normally use ASTK to run my cases, but I thought about switching (going with the time ...). First try, first issue I guess.

I too can't share a .comm file due to confidentiality...

I try to paint a picture what happened though:

[list=*]

Stage from text-file,

Only the mesh-file defined as input

Output defined in second comm-file (stage) but the error came right after finishing the run of the first stage (STAT_NON_LINE)

[/list]

Have you been able to narrow it down?

I attached a screenshot of the strange behaviour

Attached file:
screenshot.png, 524227kb

mf

Hello,

no, I could not find the reason. I removed the POURSUITE in order to get the simulation to finish. I am not sure, it could be that there was another mistake. So, no news here, sorry,

Mario.

mf

rodrigofarias
I haven't compiled a new version in a while now, SM2023 did not work entirely, SM2024 I did not manage to compile with MPI. It seems to get more complicated with every new version.
But try this: https://github.com/jcugnoni/SalomeMeca2024_Code-AsterMPI
I am not sure it works flawlessly, though.

There's also a newer Code_Aster MPI standalone here: https://hub.docker.com/u/simvia . According to this thread it is with MPI, although I haven't tested: https://forum.code-aster.org/public/d/30829-mpi-build-failure-versuion-17321/14

Mario.