[intrepid-notify] ALCF BG/P Intrepid Scheduling Policy Change Notification
Bowen Cheetah Goletz
cheetah at alcf.anl.gov
Thu Mar 25 16:10:58 CDT 2010
ALCF User Community:
In order to improve response time for large jobs and overall utilization of the
ALCF BlueGene/P machine "intrepid", the job scheduling policy will be modified
on Monday, April 5th. On that date, the following change will go into effect:
Jobs submitted to the prod queue requesting 4K or fewer nodes but greater than
six hours wall time will be restricted to rows 4 and 5.
All other job and scheduling limits remain the same, and there are no changes to
the prod-devel queue. Users should still submit their jobs to the prod or
prod-devel queues as before and no changes to submission scripts are required.
Users should see a noticeable improvement in response time for large jobs
running on 8K and 16K partitions, with such jobs starting within six hours of
submission (barring several other large jobs already running). We also expect
increased scheduler efficiency for small sub-six hour jobs. Jobs submitted to
the 'prod' queue will automatically reroute to the queue appropriate for the job
parameters.
Note that the full machine is five rows of eight 1K-node racks.
The full description of the job scheduling policy is available on the ALCF wiki:
https://wiki.alcf.anl.gov/index.php/Job_Scheduling_Policy
-ALCF Support Team
More information about the intrepid-notify
mailing list