This isn't the most straightforward solution, but as a workaround, create a bash script and run that script through npRmpi? Something like:
openmpi -np 15 -hostfile /path/to/hostfile $1
Dr. Ethan Deneault
Assistant Professor of Physics
The University of Tampa
401 W Kennedy Blvd
Tampa, FL 33606
Barnet Wagman <firstname.lastname@example.org
There have been many postings about openmpi-default-hostfile on the
list, but I haven't found one that answers my question, so I hope you
won't mind one more.
When I use mpirun, openmpi-default-hostfile does not appear to get used.
I've added three lines to the default host file:
'node0' is the local (master) host.
If I explicitly list the hostfile in the mpirun command, everything
works correctly. E.g.
mpirun -np 15 -hostfile /full/path/to/openmpi-default-hostfile hello_c
works correctly - hello_c gets run using all three nodes.
However, if I don't specify the hostfile, only the local node, node0, is
mpirun -np 15 hello_c
creates all 15 processes on node0. I was under the impression that all
machines listed in openmpi-default-hostfile should get used by default.
Is that correct?
Unfortunately I can't use the hostfile command line option. I'm going
to be using a mpi app (npRmpi) that doesn't let me pass params to
mpirun. So I need all my nodes used by default.
openmpi 1.4.3, built from source.
OS: Debian lenny (but the Debian openmpi package is NOT installed).
Installation dir: /home/omu/openmpi
The default host file has pathname
I've set two envirnmental variables to support open mpi:
Are there any other environmental variables that need to be set?
I'd appreciate any suggestions about this.
users mailing email@example.com