|Posted on Thursday, February 14, 2013 - 1:26 pm: |
Is this a bug? When NWChem 6.1.1 runs a multi-processor job it creates one 250+ MB input.anoits.N file for each processor (core). If the job is killed by WebMO, say because it exceeds a time limit, all these files are moved by WebMO from the compute node's /scratch directory to the user's directory on the webmo server.
Post Number: 334
|Posted on Thursday, February 14, 2013 - 5:55 pm: |
I am pretty sure the issue is that (by default) NWChem stores scratch files in the current directory. Thus they were probably never in /scratch on the compute node anyway, but rather in the directory WebMO was using on the compute node.
Try adding the "scratch_dir" directive right in the input file, before WebMO submits it. That fixes things in my testing. I will add this feature to the next release of WebMO. In the mean time, you can simple add this to the NWChem job templates to make this the default.
|Posted on Thursday, February 14, 2013 - 8:47 pm: |
Thanks, JR. That works! Sorry I did not pick up on that suggestion from the previous post on this issue.