code_saturne User's Forum

Hello, everyone.

I was trying to compile CS in my own directory again due to the recent library update on ARCHER.

I exactly followed the summary of the configuration options and full compile instructions at http://www.archer.ac.uk/documentation/s ... phase2.php for reference.

The initialization was fine as I have obtained a result file folder with a mesh_input in it.

But when I tried to run CS, it gave errors like:

"Rank 11 [Mon Apr 4 13:39:09 2016] [c0-0c0s1n1] Fatal error in PMPI_Alltoallv: Other MPI error, error stack:
PMPI_Alltoallv(557)............: MPI_Alltoallv(sbuf=0x14091c0, scnts=0x13d8838, sdispls=0x1409340, MPI_BYTE, rbuf=0x8c50f0, rcnts=0x13d8538, rdispls=0x1409940, MPI_BYTE, MPI_COMM_WORLD) failed
MPIR_Alltoallv_impl(380).......:
MPIDI_CRAY_ugni_alltoallv(1373):
MPIU_ugni_wait_rdma_events(412): GNI_CqGetEvent (GNI_RC_SUCCESS)
Rank 9 [Mon Apr 4 13:39:09 2016] [c0-0c0s1n1] Fatal error in PMPI_Alltoallv: Other MPI error, error stack:
PMPI_Alltoallv(557)............: MPI_Alltoallv(sbuf=0x85edd0, scnts=0x1a5a848, sdispls=0x85ef50, MPI_BYTE, rbuf=0x11dbd70, rcnts=0x1a5a548, rdispls=0x85f550, MPI_BYTE, MPI_COMM_WORLD) failed
MPIR_Alltoallv_impl(380).......:
MPIDI_CRAY_ugni_alltoallv(1373):
MPIU_ugni_wait_rdma_events(412): GNI_CqGetEvent (GNI_RC_SUCCESS)
......" (See attached for details)

I am very confused about these errors and it may have loaded an incorrect version of one of its dependencies or so.

I also found the error happened when the computation was at the step "Partitioning 13476280 cells to 192 domains on 192 ranks (SCOTCH_dgraphPart)." May be it is SCOTCH problem?

Does anyone know about this?

Sean

Hello,

This is probably an installation issue, but to check if the issue is not simply due to PT-Scotch, you can force a different partitioning scheme using the performance tuning tab in the GUI or the cs_user_performance_tuning.c user subroutines.

Regards,

Yvan