issues with MPI


My parallel simulation with MPI often stops prematurely because of a mystic segmentation fault that seemed to occur at different time steps each time I tried running the same simulation. On the other hand, the serial version has no such problem. I was using OpenMPI/intel/3.0.1 when the problem occurred. I then tried a newer version, OpenMPI/intel/4.0.1, and the problem seemed to go away. I am thus curious to know if there is a particular version of MPI that works the best for CROCO. Also, the cluster computer I am using has the PGI compiler in addition to the intel one I am using now. I am wondering if there is any preference between those two compilers for CROCO.


Hi Guangyu

There shouldn’t be any dependency to the version of the MPI version, the way it is implemented is pretty standard. About PGI, there is no preference a priori, I think it could worth trying to check which one is faster in your case.

Thanks for your answer, Rachid. I think the difference I saw between the two versions (3.0.1 vs 3.1.3) of MPI was likely caused by some issue with their installations on the cluster I am using rather a general problem with the CROCO source code