[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Condor-users] mcnp5 and condor -- file not found, unit 39?
- Date: Mon, 14 May 2007 17:06:34 -0500
- From: Todd Tannenbaum <tannenba@xxxxxxxxxxx>
- Subject: Re: [Condor-users] mcnp5 and condor -- file not found, unit 39?
Grant Goodyear wrote:
I sent the attached message to the mcnp forums, but after a bit
of digging I'm wondering if my file-not-found error is due to
the fact that mcnp uses unnamed fortran scratch files, which it might be
trying to write where it doesn't have permission. I'm using
condor-6.9.2, and I can generate a similar (if not identical) error
with a simple hello-world fortran program that writes to a scratch
file. Is there a canonical fix for this problem that avoids having
to hack and recompile mcnp (which would be highly non-trivial, since
I don't have a fortran-90 compiler for windows)?
Do you have "control" over your execution machines?
If so, find out what directory mcnp is trying to write its scratch files
a) change the permissions of the directory to allow the execute account
(aka the one Condor is using to run jobs) access to that directory.
b) tell Condor to run jobs on Windows using a different account; one
that already has permission to write to the directories in question.
you can either put the name of the account to use in the condor_config
file, or you can tell Condor to execute the job as the same user who
submitted the job.
Another possible idea:
Perhaps your fortran runtime library allow you to configure the location
where scratch files are created based upon an environment variable? Or
perhaps it looks at the "TEMP" environment variable? If so, then you
can set it to the Condor execute directory (i.e. the subdirectory your
job is born in) and you are all set.