Hi Again !

Dirk Eddelbuettel wrote:


Works for me (though I prefer salloc), suggesting that you did something to
your network topology or Open MPI configuration:

:~$ cat /tmp/jerome_hw.c
// mpicc -o phello phello.c
// mpirun -np 5 phello

#include <unistd.h>
#include <stdio.h>
#include <mpi.h>

int main(int narg, char *args[]){
  int rank,size;
        char ProcessorName[MPI_MAX_PROCESSOR_NAME];
        int ProcessorNameLength;

  MPI_Init(&narg,&args);

  MPI_Comm_rank(MPI_COMM_WORLD,&rank);
  MPI_Comm_size(MPI_COMM_WORLD,&size);

        MPI_Get_processor_name(ProcessorName,&ProcessorNameLength);
        sleep(11);
        fprintf(stdout,
                        "Hello world! I am %d of %d and my name is `%s'\n",
                rank,size,
                ProcessorName);

  MPI_Finalize();

  return 0; }

//
// End of file `phello.c'.

:~$ mpicc.openmpi -o /tmp/jerome_hw /tmp/jerome_hw.c
:~$ orterun -np 2 /tmp/jerome_hw
Hello world! I am 1 of 2 and my name is `xyz-1'
Hello world! I am 0 of 2 and my name is `xyz-1'
:~$ salloc orterun -np 2 /tmp/jerome_hw
salloc: Granted job allocation 421
Hello world! I am 0 of 2 and my name is `xyz-1'
Hello world! I am 1 of 2 and my name is `xyz-1'
salloc: Relinquishing job allocation 421
:~$


The above submission works the same on my clusters.
But in fact, my issue involve interconnection between the nodes of the clusters:
in the above examples involve no connection between nodes.

My cluster is a cluster of quadcore computers:
if in the sbatch script

#SBATCH --nodes=7
#SBATCH --ntasks=15

is replaced by

#SBATCH --nodes=1
#SBATCH --ntasks=4

everything is fine as no interconnection is involved.

Can you test the inconnection part of the story ?


| I have set no MCA parameter, and the firewalls are off, and the kernels 
(2.6.16) were built with no Security feature.

Try simplifying further: no default hosts beside localhost etc.  Try orterun
before you try salloc.  Simplicity first.

I try to keep thing simple (and secure):
I will double check my set up.

Thanks,
Jerome




Dirk

Reply via email to