...
We have found that submitting many 1 cpu jobs to the clusters can often create efficiency problems when these jobs compete for resources with jobs that are requesting whole or multiples of nodes. That is, the 1 cpu jobs will dominate the resources making jobs requesting multiples of nodes wait longer than they should with respect to what they should using the fair share policy. To overcome this difficulty we suggest that your 1 cpu jobs should be submitted in batches according to the size of the node. To achieve this we have installed a parallel job manager using the perl program packageprogramming language. To utilize this functionality you will have to write a perl script to manage your 1 cpu jobs. Below we give an example of what your scripts might look. In Example 1 is a normal 1 cpu PBS script executing your program without any perl parallelization. Examples 2a and 2b show the pbs script and perl script that you would need to write in order to parallelize example 1.
...