Computational Cluster Programs

How to Run GaussView on ATS-Hosted Clusters

Note that, by contractual agreement with Gaussian, Inc., only authorized members of the UCLA community can use Gaussian. Inquiries should be directed to Ashay Patel, UCLA Department of Chemistry or to atshpc@ucla.edu. The consultants will reply during normal business hours.

How to Run GaussView in Batch

GaussView is not available in batch. See How to Run Gaussian on ATS-Hosted Clusters for information about running Gaussian in batch.

How to Run GaussView Interactively

To Run GaussView from the UCLA Grid Portal

To run GaussView from the UCLA Grid Portal click the Interactive tab. Then click Xterm. If you have access to a cluster that has GaussView installed and you are on the access list for Gaussian, at the shell prompt, enter:

gaussview

To Run GaussView from the Cluster Login Node

To run GaussView interactively you must first connect to the cluster login node with X11 forwarding enabled. Then use ssh to access one of the interactive nodes also with X11 forwarding enabled -- see Computing on the Hoffman2 Cluster: Running Interactive Programs. Or use qrsh to access one of the compute nodes interactively -- see How to Get an Interactive Session through SGE. You must use qrsh if you want to run Gaussian from within GaussView and use more than one processor.

Then, at the shell prompt, enter:

gaussview

How to Run Gaussian from within GaussView

The GaussView program will let you submit a com file to Gaussian. The com file may be one created by GaussView, or one you created with a text editor. When you close your GaussView session, any Gaussian process that is still running will be aborted. If your calculations need to run for an extended period of time, we recommend that you instead run Gaussian in batch.

If you want to run Gaussian in serial mode, you do not need to know any of the following information.

There are three possible methods of running Gaussian within GaussView: serial, multi-threaded (SMP-parallelism), and parallel (Linda-parallelism). See "Using Gaussian 09 with Linda" http://www.gaussian.com/g_tech/g_ur/m_linda.htm for more information about running Gaussian multi-threaded or parallel. Note that not all Gaussian calculations are parallelized and so not all calculations will benefit from running on multiple nodes. Linda-parallelism is not available with Gaussian 03 on the Hoffman2 Cluster.

The %NProcShared and %LindaWorkers instructions in Gaussian 09 determine if Gaussian will run multi-threaded, or parallel, or both multi-threaded and parallel. Either add one or both instruction lines at the top of your com file, or use the Calculate=>Gaussian Calculation Setup dialog Link 0 panel to supply the same information. See Figure 71: The Link 0 Panel, in http://www.gaussian.com/g_tech/gv5ref/jobs.htm for more information.

You can edit your com file either before you start GaussView, or from within your GaussView session with the Calculate=>Gaussian Calculation Setup dialog Edit button. Or you can supply %NProcShared and %LindaWorkers values with its Link 0 panel. Here are some examples that use qrsh for each scenario.

Serial (single processor, single node)

To run Gaussian in serial mode, at the cluster login node shell prompt, enter:

qrsh -l mem=1024M,time=8:00:00

Then, at the compute node's shell prompt, enter:

gaussview

No "-pe" parallel environment directive is used. No %NProcShared instruction should be present in your com file. No %LindaWorkers instruction is needed in your com file.

Multi-Threaded (2 to 8 processors, single node)

To run Gaussian multi-threaded, at the cluster login node shell prompt, enter:

qrsh -pe shared 4 -l mem=1024M,time=8:00:00

or,

qrsh -pe shared_msa 4 -l mem=4096M,time=8:00:00

Use the shared parallel environment to run on IDRE Data Center nodes. Use the shared_msa parallel environment to run on Math Science Data Center nodes.

Then, at the compute node's shell prompt, enter:

gaussview

The "-pe shared 4" or "-pe shared_msa 4" directive requests a total of 4 processors for your qrsh session, and that they all be on the same node. Add at the beginning of your com file or with the Link 0 panel:

%NProcShared=4

Note that the %NProcShared value is processors per node. No %LindaWorkers instruction is needed in your com file.

Parallel (more than 8 processors, more than one node)

To run Gaussian in parallel, at the cluster login node shell prompt, enter:

qrsh -pe 4threads* 16 -l i,mem=1024M,time=8:00:00

The "-pe 4threads* 16" directive requests a total of 16 processors for your qrsh session. They will probably be on 4 different nodes, with 4 processors per node. To find out which nodes and how many processors per node have been assigned to your qrsh session, after SGE has started your qrsh session, at the compute node's shell prompt, enter:

get_pe_hostfile

The get_pe_hostfile script will return some information like:

a2 4 inter_idre.q@a2 <NULL>
a3 4 inter_idre.q@a3 <NULL>
n26 4 inter_idre.q@n26 <NULL>
n27 4 inter_idre.q@n27 <NULL>

Then, at the compute node's shell prompt, enter:

gaussview

Add at the beginning of your com file or with the Link 0 panel:

%NProcShared=4
%LindaWorkers=a2,a3,n26,n27

Note that the %NProcShared value is processors per node. The node names you supply to the %LindaWorkers instruction will be different for each qrsh session. Linda-parallelism is not available with Gaussian 03 on the Hoffman2 Cluster.

November 2009