working directory

[Glen Beane]

I'm starting a job with mpiexec, but the working directory on all the
compute nodes is not correct.

Here is the situation:
The datafiles are all in one directory structure, the executable in
another.
In the pbs script(i'm using SPBS) I cd to the base directory for the data,
then I call mpiexec, and I use the full path to the executable.  The
executable starts on all the nodes, but then they all can't open their
configuration file.
If I use ssh(mpirun) the compute nodes seem to be in the correct working
directory
- they open the configuration file, then their specific block of data.

I though mpiexec would start the executable in the same direcory it was
called from, so if I changed to the data directory in the PBS script and
then call mpiexec, I thought that that would be the working directory for
every process - which doesn't appear to be the case.  

Does anyone know how I can solve this problem?  If I use ssh to start the
job it takes about 15 minutes(400 processor job), so I really want to be
able to use mpiexec.

Glen Beane
- University of Maine Supercomputing
- University of Maine Department of Computer Science