Sun Grid Engine on DCIS research cluster

Overview

Sun's Grid Engine is now installed on the IBM Linux machines in the DCIS research cluster. It is the basic default installation with a single queue defined on each machine. The queues have no restrictions on memory or CPU-time, nor are there access restrictions. Anyone with a valid account on the research cluster can submit jobs.

Usage

In order to use SGE, [t]csh users should put the following in your .login:

    source /res/users/sgeadmin/default/common/settings.csh

Similarly, sh users should put the following in your .profile:

    . /res/users/sgeadmin/default/common/settings.sh

Simple example

I tried running the example job, simple.sh, Sun supplies and discovered an interaction with my .login file. I was able to avoid this by putting the line

    if ($?SGE_STDOUT_PATH) exit

as the first line of my .login file. (I later discovered this interaction was due to the file being named simple.sh. Files named with shell names are invoked with the full login environment. In the case of [t]csh users, that means sourcing both the .cshrc and .login files.

Shell scripts are submitted using the qsub command:

    qsub filename

You can check on the status of running jobs with the qstat command. Output from jobs are returned in filename.on (for standard out) and filename.en (for standard error). (n is the job ID assigned by SGE and displayed by qstat while the job is running.)

Current status

The current status of the queues on the SGE can be viewed with the qstat [-f] command. I update a page summarizing the SGE status with links to its load average history online.

Note:

The qlogin and qrsh commands are based on telnet and rsh. Since these are no longer enabled on our machines, these features do not currently work. If someone has a use for interactive features under SGE, email to help@research and I will look into configuring it to work with ssh.

For further information, see the User's Guide.

This page last updated March 9, 2011.