problems in mpi initialization

[simonecamarri]

Dear Mikael,

I have the following problem in running a first simulation with Oasis:

simone@sc:Oasis$ python NSfracStep.py problem=DrivenCavity solver=IPCS_ABCN
*** The MPI_Allreduce() function was called before MPI_INIT was invoked.
*** This is disallowed by the MPI standard.
*** Your MPI job will now abort.
[sc.local:42195] Local abort before MPI_INIT completed successfully; not able to aggregate error messages, and not able to guarantee that all other processes were killed!

It seems this is a problem in mpi initialization, probably intrinsic in Fenics. I have installed pyMPI but the problem persists. Do you have any suggestion?

Best regards,

Simone

[mikaem]

Hi Simone,

I have seen that error message before on the Scinet supercomputer at the University of Toronto. It's a dolfin bug of some sort if I remember correctly and it was "fixed" by creating a mesh as the first thing after doing from dolfin import *. See top section of problems/init.py, uncheck line number 12. If you are using more than 400 cpus, then change the size of the mesh.

Mikael