Conversion of LSF Batch Farm servers to RHEL5 64-bit.

 

As noted in the comp-change posting of October 2, 2009 (http://www.slac.stanford.edu/cgi-bin/lwgate/COMP-CHANGE/archives/comp-change.200910/date/article-0.html) we have begun converting the Computing Division batch farm servers to 64-bit Red Hat Enterprise Linux 5 (RHEL5).  The RHEL5 systems are available in the LSF general queues (short, medium, long, xlong, xxl, idle) along with 32-bit RHEL3 and 64-bit RHEL4 systems.  We will not immediately convert all systems to RHEL5; however, to satisfy requirements from some of our large user groups, we need to convert a significant portion of the farm as quickly as we can.

 

Currently, of our total 2940 job slots in the general batch farm, 54% are RHEL5, 21% are RHEL4 and 25% are RHEL3.  These numbers will change in favor of RHEL5 as we continue to upgrade. By the end of November we expect the percentages to be approximately 10% RHEL3, 10% RHEL4 and 80% RHEL5.

 

If you are specifying a particular Red Hat OS version in your job submissions, such as “rhel30” or “rhel40” or “rhel50” please be aware that this limits the pool of machines your job may run on.  This submission specification may be unnecessary.

 

In nearly all cases any binary compiled on a RHEL3 or RHEL4 system will run correctly on a 64-bit RHEL5 system, so adding this OS to the batch farm should be transparent to most users. If you do run into problems running your existing jobs on a RHEL5 batch system, please send a report to unix-admin@slac.stanford.edu so we can investigate.  If necessary, you can direct your jobs to run on, or avoid, a particular operating system by specifying resource names of 'rhel30', 'rhel40' or 'rhel50'.  For example, to force your job to run on a RHEL4 system, do

 

 bsub -R rhel40 <job_script>

 

To force it to avoid the RHEL5 systems, do

 

 bsub -R “\! rhel50” <job_script>

..or..

 bsub –R “-rhel50” <job_script>

 

 

If you specify none of these resources, your job will run on any Linux machine that is available when the job is dispatched.

 

We will be migrating our entire batch farm to RHEL5 at some point so it is important that you test your code in this environment and notify us of any problems.

 

If you have any questions, please send email with an appropriate subject line to unix-admin@slac.stanford.edu.