MCSR provides two ways to run Gaussian jobs on our Beowulf Cluster, Mimosa. A normal version of Gaussian is available for running serial Gaussian jobs. A parallel version of Gaussian, compiled with LINDA, is available. However, we ask that you only use the LINDA version for solving those types of Gaussian problems that can approach linear speed-up on our cluster (e.g., Hartree-Fock, DFT, and MP2, but not CCSD.)

Previously, users had to create and modify their own PBS scripts in order to run normal Gaussian jobs harmlessly on mimosa. Otherwise, LINDA licenses would be ciphened to run these non-parallel jobs, causing problems for actual parallel jobs that might later try to run. However, we recently updated our script front-end script, g98sub, for submitting normal (serial) or parallel Gaussian jobs to PBS on Mimosa. With this new enhancement, serial Gaussian jobs for which only one processor is specified will not drain LINDA licenses, but rather automatically be directed to excute serially, without LINDA.

The format for submitting Gaussian jobs via our script is still the same:

g98sub num_nodes input_file

Please do use this script to submit your serial and parallel Gaussian jobs to mimosa, and limit the num_nodes value to the maximum number of nodes that your job can use efficiently.

We upgraded our LINDA license from 16 nodes to 64 nodes on mimosa on Friday, April 5.

New Policy: Each user may now have a total of four jobs running or queued across the mimosa PBS queues. Of these, four may be running at any given time. It is reasonable that a user could have 3 MPI jobs running and 1 queued, ready to run, or any combination of queued and running jobs adding up to 4. However, Gaussian/LINDA users will never have have any jobs in the queue. If there are not enough available nodes to run their job at the time they run the g98sub script, then the script will not submit the job to PBS. This is a necessary restriction, since the g98sub script creates an input file identifying the nodes available to run the job. If the resources are not available, the script will not submit the job. If nodes are available, but if the user already has his maximum number of jobs running, and is permitted to qeueue the next job until one of his other jobs finishes, then the one or more of the nodes originally identified by the script as being available might, in the meantime, be allocated to a subsequent job, in which case the queued job will still not be able to run. For this reason, we no longer allow a Gaussian/LINDA user to queue jobs.

More information on Gaussian 98 on Mimosa.
More information on Computationl Chemistry at MCSR.

If you have any questions about running computational chemistry jobs at MCSR, email assist@mcsr.olemiss.edu.