Hi,
I’m running PETSc code on a cluster that uses the Sun Grid Engine batch
submission system. In this system, the working directory is set as the
directory where the job script is submitted (with -cwd flag) or the user’s home
directory (without the -cwd flag).
In my job scripts on other clusters, I usually create a new directory using the
PBS_JOBID and put the output of my petsc code into that directory. With the
Sun Grid Engine, I cannot figure out how to get output to go to this directory
— it always ends up in one of the two directories mentioned above.
(If anyone knows how to fix this problem in the SGE job script, please inform
me!)
My current approach is to give petsc output filenames with the full path, e.g.
/work/katz/ss3093/ss3093_0001
Unfortunately this causes a different problem.
With PETSC_VIEWER_BINARY_MATLAB, this full path is written into the .info
file. If the output directory then gets moved to have a different path,
PetscReadBinaryMatlab.m then chokes because it looks for the binary file down
the incorrect path.
As a workaround, I have added these lines into PetscReadBinaryMatlab.m
immediately after line 35 (though I’m sure there is a better way to do this).
% check for incorrect path specified in .info file
if (~isempty(strfind(str,'PetscOpenFile')) &&
~exist(Set.filename,'file'))
dirsep = strfind(Set.filename,'/');
if dirsep; Set.filename = Set.filename(dirsep(end)+1:end); end
end
Cheers,
Rich
