Newsgroups: comp.parallel.mpi
From: "Andrew Mc.Ghee" <mcghee@mech.uq.edu.au>
Reply-To: mcghee@mech.uq.edu.au
Subject: MPI Working Directories
Organization: Mechanical Engineering, University of Queensland, Australia
Date: Sun, 25 May 1997 11:54:45 +1000
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Message-ID: <33879BE5.2399@mech.uq.edu.au>

Whenever running an MPI job with command options, we need to use the
following command,

mpirun -arch LINUX -np 2 /nova5/mcghee/bin/%a/p.x
/nova5/mcghee/cases/flatplate/plate.mb

instead of what we thought would work;

mpirun -arch LINUX -np 2 /nova5/mcghee/bin/%a/p.x plate.mb

For some reason, the MPI program p.x starts up with
	/nova5/mcghee/bin/%a/
as the working directory (issuing a system("pwd") from within the mpi
executable), instead of where the script was started;
	/nova5/mcghee/cases/flatplate/
(and the same happens when we add other architecture machines to the
list).
So we are having to specify the working directory explicitly for all
file I/O.
This isn't a major problem, just more an anoyance - but I seem to
remember other MPI environments that we have used not having this
problem. Does it require modifications to the mpirun command to stop
using the executable directory as the working directory? Or is there
something else at work here.

Thanks in advance to any advice given.

regards,
Andrew Mc.Ghee
Mechanical Engineering
University of Queensland