Dear OpenMPI user group,
in former OpenMPI release versions I have used a combination of these two
options:
--mca rmaps_dist_device
--map-by dist:span
in order to let OpenMPI automatically map the processes close to the network
card. This was useful for PingPong benchmarks using only one process
I'm having problems running programs compiled against the OpenMPI 5.0.1
package provided by homebrew on MacOS (arm) 12.6.1.
When running a Fortran test program that simply calls MPI_init followed by
MPI_Finalize, I get the following output:
$ mpirun -n 2 ./mpi_init_test
--
OMPI seems unable to create a communication medium between your processes.
There are few known issues on OSX, please read
https://github.com/open-mpi/ompi/issues/12273 for more info.
Can you provide the header of the ompi_info command. What I'm interested on
is the part about `Configure command li
Thanks, George, that issue you linked certainly looks potentially related.
Output from ompi_info:
Package: Open MPI brew@Monterey-arm64.local Distribution
Open MPI: 5.0.1
Open MPI repo revision: v5.0.1
Open MPI release date: Dec 20, 2023
MPI
Adding '--pmixmca ptl_tcp_if_include lo0' to the mpirun argument list seems
to fix (or at least work around) the problem.
On Mon, Feb 5, 2024 at 1:49 PM John Haiducek wrote:
> Thanks, George, that issue you linked certainly looks potentially related.
>
> Output from ompi_info:
>
>
That would be something @Ralph Castain needs to be
looking at as he declared in a previous discussion that `lo` was the
default for PMIX and we now have 2 reports stating otherwise.
George.
On Mon, Feb 5, 2024 at 3:15 PM John Haiducek wrote:
> Adding '--pmixmca ptl_tcp_if_include lo0' to the
Stupid question... Why is it going 'out' to the loopback address? Is shared
memory not being used these days?
On Mon, Feb 5, 2024, 8:31 PM John Haiducek via users <
users@lists.open-mpi.org> wrote:
> Adding '--pmixmca ptl_tcp_if_include lo0' to the mpirun argument list
> seems to fix (or at least
That's not for the MPI communications but for the process management part
(PRRTE/PMIX). If forcing the PTL to `lo` worked it mostly indicates that
the shared memory in OMPI was able to be set up correctly.
George.
On Mon, Feb 5, 2024 at 3:47 PM John Hearns wrote:
> Stupid question... Why is
