Dan, Using os.environ instead of tempfile seems to be the correct incantation. I also had to eliminate setting the $TMPDIR environment variable in the PBS startup script because MPI warns against using an external filesystem like Lustre. Sometimes using the Lustre filesystem worked, but it was sporadic.
So I'm guessing that now MPI gets it's local /tmp files working on each node before the python sets the $TMPDIR and now both seem happy. I have now made several runs across 4 nodes! Thanks, Bill On Apr 7, 2014, at 11:55 AM, Daniel Wheeler <[email protected]> wrote: > On Thu, Apr 3, 2014 at 1:08 PM, Seufzer, William J. (LARC-D307) > <[email protected]> wrote: >> Dan, > >> Do you mean to set the environment variable with os.environ? Would this be >> different from setting the variable as part of the PBS start up script? > > Yes. > >> Are you suggesting that all cores are creating a .msh file with the same >> name in the same place and writing over each other? > > No, I don't think that is happening. That would be very stupid. > >> I'll attempt, next week, to have the .msh ready and use Gmsh3D to read >> directly. If the .msh file is read directly is gmsh invoked? > > No. If the .msh file is already available, the Gmsh3D class shouldn't > make any more calls to Gmsh. > > -- > Daniel Wheeler > _______________________________________________ > fipy mailing list > [email protected] > http://www.ctcms.nist.gov/fipy > [ NIST internal ONLY: https://email.nist.gov/mailman/listinfo/fipy ] _______________________________________________ fipy mailing list [email protected] http://www.ctcms.nist.gov/fipy [ NIST internal ONLY: https://email.nist.gov/mailman/listinfo/fipy ]
