11. marts 2009 15:37
I need to optimize the task distribution on the Windows Compute Cluster. Is possible to set something like a backfill but between jobs? Occurs that if a job reserve 30 CPUs for 31 tasks that demands only 1 processor each, until the last task ends, it keep reserves of 30 CPUs. Ie, 29 CPUs are idle when the last task is running.
There is some registry key that allows this kind of backfill?
There is any procedure that I have forgotten?
- Flyttet af Alex SuttonOwner 26. marts 2009 01:36 moving to job scheduler area (Moved from Windows HPC Server Developers - General to Windows HPC Server Job Submission and Scheduling)
26. marts 2009 14:41RedaktørThat is the expected behavior; jobs hold whatever resources they start with for their entire life. The only way around this in CCS is to submit each of your tasks as a separate job.
FWIW, in HPC Server 2008 there is a new policy called Grow and Shrink Scheduling which does exactly what you are asking for.
- Markeret som svar af Josh BarnardModerator 26. marts 2009 15:15