[Rojj]

Hi,

Follow-up to Is it possible to avoid reconstructParMesh?


I am now trying to post-process (postProcess -func ....) on the decomposed case with

mpirun --mca orte_base_help_aggregate 0 -np X postProcess -internalCloud -parallel

internalCloud has around 5,000 points and the mesh is around 29,000,000 cells.

It seems that it is taking far too long. To give you a reference:

• Cores = 36 (AWS Ubuntu)
• Sim time = 6.5 hours
• postProcess = 2.5 hours (so far!!!)

I have checked and all cores are at 100% and the stdout of `postProcess`
is stuck at Create mesh for time XXXX

Do I need to reconstruct the mesh before sampling?

Any suggestion? Thanks!

[Rojj]

I have been doing some more testing. For some reasons I cannot run postProcess internalCloud on large cases.

This was my initial approach. On a decomposed and converged case

mpirun -np X postProcess -func "mag(U)" -latestTime -parallel
mpirun -np X postProcess -func internalCloud -latestTime -parallel

The first calculation completes without problems, but internalCloud hangs.

This works instead

mpirun -np X postProcess -func "mag(U)" -latestTime -parallel
reconstructPar -latestTime
postProcess -func internalCloud -latestTime

But of course this is serial.

It works, but I don't really understand why.