[Carpet] Cactus and Openmpi

[Hee Il Kim]

Hi,

Now I'm testing a new cluster with openmpi-1.2.5. The cluster is composed of
10 nodes of 2 quad core Intel Harpertown cpus. Simple mpi_hello test passed
and there's no problem if I take only one node. But if I take 2 nodes, then
it shows

...
...

              CartGrid3D: Apply symmetry boundary conditions
            Boundary: Unselect all grid variables for boundary conditions
        Whisky: Convert back to primitive variables (polytype)
--------------------------------------------------------------------------------
[compute-0-1.local:20975] *** An error occurred in MPI_Allgather
[compute-0-1.local:20971] *** An error occurred in MPI_Allgather
[compute-0-1.local:20971] *** on communicator MPI_COMM_WORLD
[compute-0-1.local:20971] *** MPI_ERR_INTERN: internal error
[compute-0-1.local:20971] *** MPI_ERRORS_ARE_FATAL (goodbye)
...
...

I'm trying to use this openmpi because the scalability with mpich2/mpich was
bad. Any hints would be appreciated.

Hee Il
-------------- next part --------------
An HTML attachment was scrubbed...
URL: /archives/developers/attachments/20080319/96674785/attachment-0001.htm