How To launch Guassian jobs at VPAC on trifid
Launching Gaussian03 jobs at VPAC on the trifid cluster is straightforward, requiring just changing a few lines in a pbs script and reflecting those changes in your input files.
An example job can be found at /usr/local/examples/gaussian To run the example job, first copy the job to your home directory on trifid.
cp -r /common/examples/gaussian .
(note there is a space and then a dot at the end of that second line.)
This job is a simple geometry optimization of a molecule running on 4 cpus. To launch the job, change into the job directory and launch with qsub.
For each job you launch, it is good practice to create separate folders and launch the job withing that folder. The example job is running on 4 CPUs. To change that to 2 CPUs change the "#PBS -l nodes" line to read:
#PBS -l nodes=1:ppn=2
make sure to also your input file to reflect the changes in the first few lines:
On trifid, your Gaussian jobs may have up to 8 cpus, and 4GB of memory is the default for each cpu. So for nproc=2 mem = 2x4= 8GB, for nproc=4 mem = 4x4=16GB, etc.
The pbs_script_gaussian copies the relevant files to the /tmp directory for better Gaussian performance and then copies the data back to the users job directory at the completion of the job.
IMPORTANT! - make sure to have enough walltime to cover the length of your job as defined in the line:
#PBS -l walltime=24:0:0
in the pbs script. If the job takes longer than the walltime data may be lost! If you think you need more walltime after you have already submitted a job, please email email@example.com and quote the job number.
When running a PBS script you need to set up your GAUSS_SCRDIR (gaussian temp directory) variable to point to somewhere in your /lustre directory.
As all Trifid gaussian users issue "module load gaussian" they will get a brief message reminding them to set this variable .. it really has become that important.
Note: The "export" line in your pbs scripts needs to be AFTER the "module load gaussian" line, as the gaussian module itself sets a GAUSS_SCRDIR to "/tmp" on the local node, overwriting your export.
Checkpoint and Restart
Some Gaussian jobs continue from a restart or checkpoint files. In order to continue jobs from checkpoint files make sure to define the OPT_FILES="" line in the pbs script to the name of your .chk file. ie):
In this case, the restart checkpoint file must then be located in your job directory.
Also, if you are saving a checkpoint file as defined in your input file, it is not necessary to define the path, as the checkpoint file will be copied back to your job directory. ie):
(this does not work!)
Please report any problems you have to firstname.lastname@example.org.