Are you running your job inside a torque job? If you don't, open mpi will not have a hostlist and will assume that it should launch everything on the localhost.

If you are launching inside a torque job, ensure that ompi was build with torque support properly - run

ompi_info | grep tm

If you see 1 or more tm plugins listed, ompi has torque support.

Finally, make sure you're using the right mpicc and mpirun, etc.

-jms
Sent from my PDA. No type good.

From: users-bounces@open-mpi.org <users-bounces@open-mpi.org>
To: users@open-mpi.org <users@open-mpi.org>
Sent: Thu Apr 01 02:07:08 2010
Subject: [OMPI users] mpiblast only run in one node

Hi,

I've installed Torque/Maui, Open MPI 1.4.1 and mpiBlast 1.6.0-beta1 in a linux Beowulf culster with 15 nodes (node1~15).

Open MPI, mpiBlast were installed and my home directory lies in a directory "/data" which was shared by all nodes.

Open Mpi was compiled with "--with-tm" to support Torque, and mpiBlast was compiled with

 "--with-mpi" to the directory where Open MPI installed.


When I run mpiBlast by mpirun in command line, like

node1 $ /data/open-mpi/bin/mpirun -np 34 /data/mpiblast/bin/mpiblast -p blastp -d nr -i test.faa -o test.out

I noticed all 34 mpiBlast processes run on node1 only.


This would not change if I submit job to Torque.

I've searched the mailing list archives but it seems won't help.

How to resolve this problem?

Any suggestion will be appreciated!



HZ Liu