A number of manuals for the operation and use of the cluster are available on this page.
Before using the cluster please read at least the user manual so that you know how to use the cluster effectively. If you are programming you own applications with MPI (the Message Passing Interface), please read the programmers manual as well.
Jobs on the cluster should be sumitted using the Torque/Maui scheduler running on this system. The following presentation gives you an idea of how to setup a job file and submit jobs to the scheduler. It will also show how to monitor your jobs on the cluster:
Torque is based on the OpenPBS job scheduler and it's website can be found here
. The Torque admin manual can be found here
. An extensive page on Maui commands can be found here
: If you submit multiple jobs at the same time using a for or while loop, it is very important to include a 'sleep' of at least 3 seconds before or after every qsub command. If jobs are submitted to quickly the scheduler will have problems scheduling nodes correctly and more than two jobs could potentially be scheduled on one node (which is not good because a node has only two processors).Note 2
: Because of a bug, beorun does not work with programs that read files through stdin and write to stdout. For these programs a workaround using the commands beomap and bpsh is needed. Here is an example of the usage if you need this:
BEOWULF_JOB_MAP=`beomap -np 1 -nolocal`
bpsh $BEOWULF_JOB_MAP ./siesta < gold.fdf > gold.out
The pbs_serial_job.sh example script in /usr/local/share/pbs contains these lines that need to be uncommented if needed.