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

Re: [Condor-users] priority groups and preemption

On 4/25/07, Jonathan D. Proulx <jon@xxxxxxxxxxxxx> wrote:

I'm running:

$ condor_version
$CondorVersion: 6.8.2 Oct 12 2006 $
$CondorPlatform: X86_64-LINUX_RHEL3 $

I have a set general use machines on which all users are equal and a
set of machines purchaced by a research group who understandably wants
preemptive priority access to their machines, but is willing to share
free cycles.

So I want to be sure that if a member of this Priority_Group submits a
job and there are no free resources a jon or jobs run my non group
members on their systems will be preempted, but Priority_Group member
jobs are never preempted.

Currently I'm just using a START = owner == "foo" || owner == "bar"
type statement to keep their nodes available.

It seems I can do what I want by defining
Priority_Group = owner == "foo" || owner == "bar"
and using that macro in the Rank, Suspend and Preempt sections of the
condor_config.local, but after reading and rereading the manual I'm
still not quite sure how to guarantee a non-priority job will be

Higher RANK (as in startd rank) currently guarantees preemption.

If a machine prefers to run one job over another via this
configuration* it guarantees a preemption. The only thing which can
delay this is retirement time, if you are not using the retirement
features then the jobs will be preempted.

The speed with which they are preempted is controlled by the KILL
expression (and the aforementioned retirement time)


* assumption: it is willing to START the new higher ranked job as well