[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[condor-users] preemption



Regarding previous attempts to get job specific tiered preemption verses user related preemption

Have tried a combination of:

START = True
RANK = ((TARGET.TIER =?= "Critical") * 1000) + ((TARGET.TIER =?= "High") * 100) + ((TARGET.TIER =?= "Normal") * 10) + ((TARGET.TIER =?= "Low") * 0)
PREEMPTION_REQUIREMENTS = TARGET.TIER =?= "Critical" && MY.TIER =!= "Critical" || ( ((CurrentTime - EnteredCurrentState) <=  10 * 60) && MY.RANK < TARGET.RANK ) || ( ((CurrentTime - EnteredCurrentState) >=  24 * (60 * 60))  && (RemoteUserPrio > SubmittorPrio * 1.2))
PREEMPTION_RANK = (EnteredCurrentState - MY.RANK ) + RemoteUserPrio
PRIORITY_HALFLIFE = 1
PREEMPT = False

to prevent preemption unless the job is very new (or believed to be runaway > 24 hours)

This seemed to be functioning just about right but due to one of:

1) more users are active
2) more jobs running
3) more machines are in the farm

The farm now thrashes jobs in the first 10 minutes

I therefore conclude that either:
A) MY.RANK < TARGET.RANK is not valid in the context of evaluation preemption requirements
(am trying CurrentRank instead of MY.RANK in case)
B) something else is triggering preemption I cannot control (but which respects preemption_rank so appears to obey the 10 minute rule in preemption requirements but doesn't really)

I could do with sorting it out before 6.7.1. any ideas?

Placing one of the three users jobs on hold and leaving only two users appears to prevent the problem but is not a tenable medium term solution.

Matt


*****************************************************************
Gloucester Research Limited believes the information 
provided herein is reliable. While every care has been 
taken to ensure accuracy, the information is furnished 
to the recipients with no warranty as to the completeness 
and accuracy of its contents and on condition that any 
errors or omissions shall not be made the basis for any 
claim, demand or cause for action.
*****************************************************************

Condor Support Information:
http://www.cs.wisc.edu/condor/condor-support/
To Unsubscribe, send mail to majordomo@xxxxxxxxxxx with
unsubscribe condor-users <your_email_address>