A recent customer requested that in building their cluster, they wanted the PGI compilers, of which we purchased the media and kit from PGI - The Portland Group.
Last week, after creating the master node and installing OSCAR, and cloning the nodes, I went thru the process of getting the master ready for their install script using rsh, .rhosts, and hosts.equiv. A little bumpy start, but I was able to get their package loaded, and temporary licensed. Upon rebooting I was getting some PBS errors and found the aside from installing the PGI compilers, they put down their version of the Portable Batch Scheduler in the /usr/local/pbs directory, passed it out to the nodes, and created their own startup script.
Upon rebooting the master, I get a few error messages mainly due to the fact that there is a pbs scheduler already running. One of the completion steps in OSCAR is to test the clusters which they are submitting jobs from their pbs install directory of /opt/pbs, while their pbs install directory is /usr/local/pbs. Since these two were killing each other, OSCAR does have an option not to install their pbs scheduler. That's what I did this morning. I reloaded the master, omitting the pbs install on OSCAR's side, finished up cloing the nodes with the systemimager package, and got my .rhosts and hosts.equiv ready and with the PGI env variable set to install the PGI software. All went well.
Looking at OSCAR's pbs test scripts they were referencing a pbs home directory of /opt/pbs and /opt/pbs/bin to issue the qsub jobs for testing. I thought i would get tricky and create a sym link to where they should be expecting their pbs directory to be based off their testing scripts and linked it to your /usr/local/pbs directory. But the OSCAR tests still failed.
Bottom line before I ramble on anymore, OSCAR's pbs tests fail, but I am sure that as far as the cluster and master is concerned with PGI, everything is fine with their portable batch scheduling from your side of the equation. Have you had any calls concerning this incompatibility? Should I just leave the system with PGI the way it is now, with their compilers and the PBS system as it is, and forego the fact that the pbs OSCAR tests are inconsequential to your compilers and the pbs working with the clusters the way it should? I thought setting a $PBS directory and modifying the testing scripts, Any Suggestions since the sym link didn't work?
I would really appreciate anyone's help in this matter.
Regards,
Paul Bounds
"Use The Web...It Works For You..."
P (281) 403-8500
F (281) 403-8591
E [EMAIL PROTECTED]
<mailto:[EMAIL PROTECTED]>
