MPI Initialisation Problem

[Tim Stitt]

Satish,

I figured out the problem. It seems that our implementation of mpiexec has 
some issues:

According to mpiexec documentation....

12. Only one process is launched, and mpiexec says "task 0 exited before
    completing MPI startup".

    This happens when you are using MPICH2, but have told mpiexec that
    it should use the MPICH1/P4 communication method.  Try with
    "-comm=pmi", and if that works, rebuild mpiexec using
    "--with-default-comm=pmi" for convenience.

So I executed with -comm=pmi and everything was o.k.

Sorry to bother you,

Tim.

On Saturday 16 June 2007 19:22, Satish Balay wrote:
> On Sat, 16 Jun 2007, Tim Stitt wrote:
> > Hi all,
> >
> > I am having some difficulty getting my parallel eigensolver to run over
> > multiple processes.
> >
> > When I execute my parallel code on my distributed-memory machine ( with >
> > 1 processes) I keep getting the following runtime message:
> >
> > mpiexec: Warning: task 0 exited before completing MPI startup.
> > mpiexec: Warning: task 1 exited oddly---report bug: status 0 done 0.
> >
> > MPI_Comm_Size returns 1 for #processes even though I mpiexec -n 2 or
> > higher.
> >
> > The code still runs to completion but serially.......
>
> can you reproduce this problem with a PETSc example? If so, send us
> the complete log of your build [configure.log, make_log_PETSC_ARCH]
> and the log for test examples [test_log_PETSC_ARCH] to
> petsc-maint at mcs.anl.gov
>
> Satish
>
> > I have taken the same code and ran it on my shared-memory machine with no
> > problems and all processes getting picked up so I know it is not a coding
> > problem.....the sample ksp test codes that come with my petsc (2.3.3)
> > distribution also exhibit this problem.
> >
> > So there seems to be a problem with the PETScInitialize routine with this
> > particular architecture. I have tried both pathscale and pgi compilers
> > with the same result. As far as I can see there are no noticeable
> > warnings generated during the config and make phases. Others use MPI
> > regularly on our cluster so I don't see how it could be a MPI library
> > issue.
> >
> > Any thoughts on what could be happening gratefully received.
> >
> > Regards,
> >
> > Tim.

-- 
Dr. Timothy Stitt <timothy_dot_stitt_at_ichec.ie>
HPC Application Consultant - ICHEC (www.ichec.ie)

Dublin Institute for Advanced Studies
5 Merrion Square - Dublin 2 - Ireland

+353-1-6621333 (tel) / +353-1-6621477 (fax) / +353-874195427 (mobile)