[intrepid-notify] Scheduling adjustments

Tisha Stacey tstacey at alcf.anl.gov
Wed Oct 29 13:18:20 CDT 2008


Over the next few weeks we will be unrolling some improvements to the
scheduling methods on Intrepid.  We are hoping to deliver simpler and
more efficient scheduling but want to roll these changes out slowly and
carefully.

One major change has been made on the production resource.  The
development queue now uses the same scheduling policy as Surveyor.  The
new priority calculation is based on a ratio of your queued time and
your requested wall clock time, and biases toward the smaller test jobs.

Next week, we will be switching prod-devel to be active 24x7 on the
current 512 nodes.  This is to address concerns regarding development
opportunities.  We will also be deploying a better method of scheduling
the jobs within the 3 production queues (short, medium and long).
Testing and simulation has shown these changes improve turn-around time
for the production load on Intrepid.

On the longer scale, these are incremental changes to some larger
improvements we will make over the next few months.  We will keep you
updated on the status of these changes.

We also will be implementing a new method for calculating priority in
the 500T queue.  This is to test the scheduling for a production roll-
out.  The calculation is based on a ratio of the queued time and the
requested wall-clock time and biasing toward large jobs.  Other features
will be added into this calculation as we move forward.

We appreciate your patience and welcome feedback.

Thanks,
ALCF Support Team



More information about the intrepid-notify mailing list