I think you actually have a few options:
1. If I'm reading the original mail right, I don't think you need
mpirun/mpiexec at all. When you submit a scripted job to PBS, the
script runs on the "mother superior" node -- the first node that was
allocated to you. In this, case it's your only node. Since you're
already running on the target node, there's nothing special that you
need to do to launch your jobs; you can just do something like this:
# Launch app 1
program-executable <inputfile1 >outputfile1 &
# Launch app 2
program-executable <inputfile2 >outputfile2 &
# Wait for both to finish
wait
Note that the following is *not* sufficient, unless you know for
absolutely sure that the 2nd executable will take longer than the first:
# Launch app 1
program-executable <inputfile1 >outputfile1 &
# Launch and wait for app 2
program-executable <inputfile2 >outputfile2
If the first app keeps running beyond the second, your script will
end and PBS will kill the job (including the first app).
2. If you really want to use mpiexec, you can use a single command
line to launch both executables, but note that we currently have
mpiexec hard-coded to only redirect stdin to the *first* executable
(this may change in future releases, but that's the way it is right
now). So you'd probably want to change your executables to read
from / write to a file directly instead of stdin/stdout.
Incidentally, using a single command line is the *only* way to get
OMPI to be aware of multiple executables and therefore launch in a
non-oversubscribed situation. This is not really much of a factor
since you're only running on a single node; it would be more of a
factor if you had grabbed a bunch of nodes and were trying to launch
individual jobs across them.
But for completeness, you could do the following (subject to the
stdin / stdout limitation listed above):
mpirun -np 1 executable1 [argv1] : -np 1 executable2 [argv2]
Hope that helps!
On Apr 24, 2007, at 9:36 AM, Edmund Sumbar wrote:
John Borchardt wrote:
I was hoping someone could help me with the following situation.
I have a
program which has no MPI support that I'd like to run "in
parallel" by
running a portion of my total task on N CPUs of a PBS/Maui/Open-MPI
cluster. (The algorithm is such that there is no real need for
MPI, I am
just as well-off running N processes on N CPUs as I would be
adding MPI
support to my program and then running on N CPUs.)
So it's easy enough to set up a Perl script to submit N jobs to
the queue to
run on N nodes. But, my cluster has two CPUs per node, and I am not
RAM-limited, so I'd like to run two serial jobs per node, one on
each node
CPU. From what my admin tells me, I must use the mpiexec command
to run my
program so that the scheduler knows to run my program on the nodes
which it
has assigned to me.
In my PBS script (this is one of N/2 similar scripts),
#!/bin/bash
#PBS -l nodes=1:ppn=2
#PBS -l walltime=1:00:00:00
mpiexec -pernode program-executable<inputfile1>outputfile1
mpiexec -pernode program-executable<inputfile2>outputfile2
Hi,
To run two serial programs on one dual-cpu node, try
#!/bin/bash
#PBS -l nodes=1:ppn=2
#PBS -l walltime=1:00:00:00
program-executable<inputfile1>outputfile1 &
program-executable<inputfile2>outputfile2
wait
--
Ed[mund [Sumbar]]
_______________________________________________
users mailing list
us...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users
--
Jeff Squyres
Cisco Systems
