mpiexec over infiniband?

[Pete Wyckoff]

charles at asc.edu wrote on Wed, 15 Sep 2004 13:05 -0500:
> 	Im trying to get mpiexec run on my Atlix350 with IB
> 
> I have compiled mpiexec with ./configure --with-pbs=/usr/pbs 
> --with-default-comm=mpich-ib
> 
> when I run a job through pbs I get the output below;
> 
> I don't really have a clue as to what that means.
[..]
> altix:/opt/asn/doc/mpi-example-ib $ cat hello-voltaire..e964
> mpiexec: Warning: read_ib_startup_ports: protocol version 0 not known, but 
> might still work.
> mpiexec: Error: read_ib_startup_ports: rank 40 out of bounds [0..1).
> read: Connection reset by peer
> altix:/opt/asn/doc/mpi-example-ib $ cat hello-voltaire..o964
> resolve_exe: using absolute exe "/opt/asn/doc/mpi-example-ib/hello-voltaire"
> node  0: name = altix3, mpname = altix3, cpu = 8
> wait_one_task_start: evt = 2, task 0 host altix3
> All 1 task started.
> read_ib_startup_ports: waiting for checkins

It means I've been slow about releasing a new version of mpiexec to
support this version of the MPI libraries.  It exists and is tested in
the CVS mentioned at www.osc.edu/~pw/mpiexec/ if you want to grab the
source directly from there.  I have plans "real soon now" to release a
tarball, too.  Sign yourself up to the mailing list to get a notice when
the time comes.

You may want to try the new mvapich-0.9.4 sometime too.  Or you can
wait until Voltaire incorporates it into their next release.

		-- Pete