在 2011-11-08二的 10:44 +0100,Tibor Pausz写道:
> Dear Moe
> 
> Ok, I see. I have to explain in detail. We have a user on our cluster,
> who wants to run for each MPI rank a special wrapper script, which setup
> things on a per MPI rank basis. He has used this technique with an other
> MPI implementation in combination with mpirun for some time.
> 
> The sbatch script is run by: "sbatch sbatch_script.sh"
> 
> My sbatch test script "sbatch_script.sh":
> #!/bin/sh
> #SBATCH --ntasks=12
> # ... other sbatch options
> srun -n 12 wrapper_script.sh
> 
> 
> My "wrapper_script.sh":
> #!/bin/sh
> # ... some stuff the user wants to run
> my_mpi_binary
> 
> So, sbatch run the sbatch_script, for each
> 
> If I start "my_mpi_binary" directly via the srun command I have no
> problems. Anything is fine. But if I use the wrapper_script I get
> defunct processes. Is this a bug, or is this beyond the MPI specification?
> 

Could you please describe your problem in detail? I think that it should
work fine with the wrapper script. 


> Best regards,
> Tibor
> 
> Am 07.11.2011 17:12, schrieb Moe Jette:
> > I suspect that he really wants to do something this:
> >
> > sbatch -n12 ../wrapper_skript.sh
> >   OR
> > salloc -n12 ../wrapper_skript.sh
> >
> > wrapper_skript.sh:
> > #setup
> > srun ./my_mpi_app
> > #cleanup
> >
> > Quoting Tibor Pausz <[email protected]>:
> >
> >> Dear all,
> >>
> >> i want to start a programm via a wrapper script. If I try this, MPI does
> >> not work correct. Is their any solution to this problem?
> >>
> >> example:
> >> srun -n 12 -l ../wrapper_skript.sh
> >>
> >> wrapper_skript.sh:
> >> #setup
> >> ./my_mpi_app
> >> #cleanup
> >>
> >> Best regards,
> >> Tibor
> >>
> >
> >
> >
> 

Attachment: signature.asc
Description: This is a digitally signed message part

Reply via email to