| | 5 | (21/09/16) ADA '''IBM MPI''' library has an unidentified issue with '''forcesoil'''. It is necessary to move to Intel MPI library (libIGCM ins_job scripts allows it). When a simulation is launched with more than 32 procs, the upper 32 range must wait for the next MPI call. The lower 32 keep running as usual. In the end, MPI calls are desynchronised. A very long run might create a MPI timeout. So the simulation crashes. In order to make sure it's not Forcesoil fault, it was tested on Curie HPC and ADA Intel MPI. No issues shown. |
