From: "Matt Challacombe (T-12)" <mchalla@t12.lanl.gov>
Newsgroups: comp.parallel.mpi
Subject: MPI and DQS
Date: Wed, 30 Sep 1998 17:09:04 -0600
Organization: Los Alamos National Laboratory
Message-Id: <3612BA10.3A7C71B0@t12.lanl.gov>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Cc: mchalla@t12.lanl.gov, extreme-linux@acl.lanl.gov, chemistry@osc.edu,
    dqs@scri.fsu.edu


Hi,

I have the following DQS and MPI difficulties I hope ya'll
can help me with (Linux 2.0.35, MPICH 1.1.1, and DQS 3.2.3).

Firstly, I have 8 dual nodes. There are two clusters, a PARALLEL
cluster with queues that are superior to those in a SERIAL cluster.

Problem 1.  If I specify 4 processors, eg.

#$ -l qty.eq.4.and.PARALLEL

but tell mpirun to use 8 processors (to take advantage of the SMP)
as for example,

mpirun -nolocal -np 8 cpi > cpi.out

I find that the run has taken place on all 8 nodes!   Is there
a way to get the machines allocated by DQS, so that I can pass it
to mpirun as a machinefile?  Is there another fix?

Problem 2.  If there are SERIAL jobs running on each node, and I
run an MPI job using the queues in the PARALLEL cluster,  the node
acting as the MPI_HOST will suspend the SERIAL job on that node,
but the rest of the active PARALLEL queues do not suspend the
SERIAL jobs that are running.

Cheers, Matt

+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Matt Challacombe
Los Alamos National Laboratory    http://www.t12.lanl.gov/~mchalla/
Theoretical Division              email: mchalla@t12.lanl.gov
Group T-12, Mail Stop B268        phone:   (505) 665-5905
Los Alamos, New Mexico  87545     fax:     (505) 665-3909
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+