From: David Turner (dpturner_at_lbl_dot_gov)
Date: 09/20/2005
Greetings Seaborg User,
You may notice Seaborg batch jobs starting out-of-order today and
tomorrow morning, in advance of tomorrow's scheduled system maintenance.
This note provides some details of the expected scheduling performance.
Seaborg is scheduled for system maintenance from 14:00 Wednesday until
03:30 Thursday (all times Pacific Daylight Time). We have configured
LoadLeveler so that it will not start any new job that cannot finish
before 14:00 tomorrow (based on the job's requested wall time limit).
As I write this, it is about 28.5 hours until the maintenance. Any currently
queued job that is requesting more than 28.5 hours will not be considered
for scheduling, regardless of when it was submitted. As we get closer
to the maintenance time, shorter jobs will be started as nodes become
available.
This scheduling behavior replaces our previous method of "draining" the
machine before maintenance periods. Note that it makes the use of the
"backfill" submit class unnecessary. We will NOT be enabling the backfill
class before tomorrow's downtime.
We prefer to checkpoint/restart the entire production workload around
these types of downtimes, but the nature of the software upgrades to be
installed in this case precludes this approach.
If you have any questions, please contact NERSC consulting at:
1-800-66-NERSC, menu option 3, 8 am - 5 pm, Pacific time
(510) 486-8600, menu option 3, 8 am - 5 pm, Pacific time
consult_at_nersc_dot_gov
http://help.nersc.gov/
--
Best regards,
David Turner
User Services Group email: dpturner_at_lbl_dot_gov
NERSC Division phone: (510) 486-4027
Lawrence Berkeley Lab fax: (510) 486-4316