Bug with open MPI

[Alexandre Guilloux]

Hello,

I have some problems to run open mpi library on my computer.

This is my configuration:

Ubuntu 10.04
Noyau linux 2.6.32-21 generic
Gnome 2.30.0

2 processors (8 cores to each one) AMD Opteron (16Go)

I have test two versions of open mpi. I have the same problem with the 2 versions :  V1.4.4 et V1.4.1

This is my procedure to install open MPI: (In my installation repertory)
- mkdir build
- cd build
- ../configure --prefix=/my_intallation_repertory
- make -j 16 all
- make install

And I use a Hello world program to test the library:

Code: Select all

/* C Example */
#include <stdio.h>
#include <mpi.h>
int main (argc, argv)
    int argc;
    char *argv[];
{
   int rank, size;

   MPI_Init (&argc, &argv); /* starts MPI */
   MPI_Comm_rank (MPI_COMM_WORLD, &rank); /* get current process id */
   MPI_Comm_size (MPI_COMM_WORLD, &size); /* get number of processes */
   printf( "Hello world from process %d of %d\n", rank, size );
   MPI_Finalize();   return 0;
}

Compil with : /my_intallation_repertory/mpicc hello.c -o hello
run with : /my_intallation_repertory/mpiexec -np 16 hello

The result is random. Some time it's work and I have this output:
"Hello world from process 0 of 16"
"Hello world from process 3 of 16"
"Hello world from process 2 of 16"
...

For the 16 process

And some time, there is a bug and I have this output:

"mpiexec noticed that process rank 10 with PID2215 on node ORDI01 exited on signal 11 (segmentation fault)"

Then I need to reboot my computer.
 
I have post my problem on the open mpi website but I don't have answer.
Has somebody have this problem? I don't know if it's a material problem, an OS problem, my procedure to install and run open mpi, ...  And I don't understand the random phenomenon.

Thanks

Best regards

Alexandre

[Yvan Fournier]

Hello,
I have never encountered this issue (including with versions of OpenMPI I built and installed similarly to what you did) but the simplest option might be to remove your Open MPI installation (using "make uninstall"), and simply install the openmpi (and openmpi-dev) package from your Ubuntu version.
Best regards,
  Yvan

[Alexandre Guilloux]

Hello,
Thanks again for your help.
I have uninstalled all my librairies with the command " make uninstall"
And I "simply install the openmpi (and openmpi-dev) package from your Ubuntu version", and now it's working...
I think I had to much version.
I have just a problem to run correctly mpd when I run the hello programm , but it will be fine.
 
Thank again for your help.
Best regard
Alex

[Yvan Fournier]

Hello,
mpd is used by some MPICH2 versions, not by Open MPI, so your hello world has probably been built using that (on Ubuntu, you may have both installed, including using packages, in which case mpicc and mpiexec are by default MPICH2, while mpicc.openmpi and mpiexec.openmpi are the OpenMPI versions).
In any case, recent versions of Code_Saturne are often tested in similar configurations, at least on recent Ubuntu versions, and should be able to use either MPI library, without getting mixed up between the 2, so you should not have additional issues now that things seem to work.
Best regards,
  Yvan