Please remember that all production jobs that are run on HPCVL servers and clusters must be submitted via Grid Engine. Test and workup jobs may still be run outside of Grid Engine on sflogin0 or swlogin1.
Adherence to this policy helps ensure system stablility and efficiency.
We have set a maximum runtime for test jobs that are executed on the workup nodes of 3 hours. This limit does not mean that production jobs of lesser runtime will be tolerated. Multiple simultaneous job runs or runs with a large number of processes are not admissable. The number of processors employed should be as low as possible.
Please note that we also require test jobs to be "niced" to 19, ie, to submitted with a "nice -n 19" prefix:
nice -n 19 program_name
The effect of this is that the test job will run with a low priority **if** competing applications are running with a higher priority. If there is no competition, "nice" has no effect. For more information, please check out the man pages: "man nice". Please note that the use of "nice" is **mandatory**, and that long-running test jobs with standard priority will be terminated without notice.
We need to assure that our login & workup node remains available for the purposes for which it is intended, particularly at times when the usage of our production nodes is high and as a consequence, waiting times on the Grid Engine are long.
We are aware that sometimes there is exceptional pressure to finish work, or code needs to be tested on large data sets, with a large number of processors, or for extended runtimes. In these cases, please contact us as we can try to make arrangements to accomodate.