|Posted on Wednesday, March 05, 2014 - 8:19 pm: |
I have system manager set to use 62 cores (on a machine with 64). In the Gaussian interface, I have it set to allow using between 1 and 62 cores per job as well. However, I recently noticed that half a dozen jobs were running simultaneously that were effectively using ~90 cores, thereby bogging the machine down quite a bit. Shouldn't the CPU cores option in System Manager prevent that?
|Posted on Wednesday, March 05, 2014 - 8:28 pm: |
Follow-up to my post above. Or is this a likely problem in our SGE queue? At first glance, it looks to be configured correctly too...
What exactly does the CPU cores value in System Manager control?
Looking at the outputs more closely, I see Linda in Gaussian had effectively reduced the number of processors at various points in some of the jobs... so maybe the nominal load was less than 62 cores, at least at the time when it allowed subsequent jobs to start? I don't use Gaussian that often myself, so I am not totally familiar with how it handles the parallel scaling.
Still, I would think WebMO would count up how many are nominall
Post Number: 391
|Posted on Thursday, March 06, 2014 - 2:21 pm: |
When using a queuing system the CPU core count in system manger is irrelevant. This is all totally a function of the SGE queue and SGE setup. You might need to tinker around a bit to get SGE set up properly. You can check the script and SGE options that were used to submit the SGE job. It is stored in the job directory, pbs_script.sh.