[mvapich-discuss] perl MPI simple not working with mvapich2-1.7

[Jonathan Perkins]

On Fri, Mar 16, 2012 at 01:58:15AM +0000, Niel Henriksen wrote:
> Hi,
> 
> I'm in need of ideas for troubleshooting a problem:
> 
> I'm running an MPI perl script with the Parallel::MPI::Simple module
> on a few different Inifiniband clusters.  It works fine with
> mvapich2-1.5.1 and 1.6, but I'm getting errors with 1.7.
> 
> Here is my mpirun_rsh command: mpirun_rsh -ssh -np 24 -hostfile
> $PBS_NODEFILE MV2_ON_DEMAND_THRESHOLD=8192 perl ./minimize.MPI.pl
> 
> Here is the error message: [tr030:mpispawn_0][child_handler] MPI
> process (rank: 2, pid: 847) terminated with signal 11 -> abort job
> [tr030:mpirun_rsh][wait_for_mpispawn] mpispawn_0 from node tr030
> aborted: MPI process error (1)
> 
> Any ideas about what changed between 1.6 and 1.7 to cause this?  Any
> other suggestions?

Thanks for the report.  Off the top of my head I can't think of anything
that should effect this.  The error message in this case isn't too
helpful so we're going to try to reproduce this with debugging enabled.

Is it possible for you to rebuild the perl module against our library
built with --enable-g=dbg --disable-fast options specified at configure
time?  With this build you can specify the MV2_DEBUG_SHOW_BACKTRACE
environment variable set to one.

-- 
Jonathan Perkins
http://www.cse.ohio-state.edu/~perkinjo