Thread: Best Practices for running Gaussian and NWChem under LSF HPC

Originally posted by: achristi, Thu Nov 22, 2007 10:50 pm

During the SC07 conference, many customers asked for best practices for running Gaussian and NWChem under an LSF HPC environment. Since the users are the experts with these tools, you probably know the best answer! So this post is to ask the community to share how they run these tools under LSF HPC, what works, what does not work, and what could be improved. If you can share a code fragment or script that would great.

This also could be a possible topic for a future Platform Users Conference talk for Hi-Ed/Life Sciences Platform users.
_________________
Alan

Originally posted by: CameronPurvis, Wed Dec 12, 2007 3:07 pm

At this time, I am far from an expert on Gaussian. However, here's what we're doing:

In our case we've basically got one research group with a group license. The Gaussian program is installed on a directory in our shared storage on the cluster. This allows the compute nodes to all see the shared Gaussian binary and run it over the shared storage. The Gaussian apps are all restricted by Unix group permissions, so that group is the only set that can execute the app. We just have to make sure that the new users from that department go into the proper groups.

Ideally, I'll get Gaussian installed on the individual compute-node image (provided the license allows it) with similar restriction on execution. If anyone is interested I can dig out pre-exec and post-exec scripts for Gaussian, and some bsub script examples. Most of those come from the actual users, who've been running Gaussian in other environments for a few years.