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

Re: [Condor-users] use of rank and dynamic slots

So the issue with dSlots and rank is that the negotiator only carves off a slot at a time.  There are modifications upstream(master/7.7.6) which allows the schedd to split said pSlot until it is consumed, and this would likely get the behavior you desire, provided you are not dependent on CC_LIMITS++HGQ.


From: "Patty Bragger" <braggepj@xxxxxxxxx>
To: "Condor-Users Mail List" <condor-users@xxxxxxxxxxx>
Sent: Thursday, March 29, 2012 10:11:06 PM
Subject: Re: [Condor-users] use of rank and dynamic slots

I forgot to include that I'm using a rank _expression_ in the job submission to get the jobs to land on the preferred machines...



From: condor-users-bounces@xxxxxxxxxxx [mailto:condor-users-bounces@xxxxxxxxxxx] On Behalf Of Patty Bragger
Sent: Thursday, March 29, 2012 10:34 PM
To: Condor-Users Mail List
Subject: [Condor-users] use of rank and dynamic slots


Hi everyone,

I'm trying to figure out how to get certain jobs to land on a specific group of machines and fill them up first before landing on other machines while using dynamic slots.

Let's say I have a group of 2 preferred machines and 5 other machines and let's say that each machine has 4 cores.  When I submit a cluster of 10 procs, during the first negotiation cycle the first two procs land on the preferred machines, then the next 5 land on the other machines.  Then in the next negotiation cycle the first two land again on the preferred machine and the last one lands on one of the other machines.  When in theory, eight of the 10 procs could land on the 2 preferred machines and only two would be sent off to these other nodes.

Am I going about this wrong?  Or is there a better way to get these jobs to more strongly prefer to land on these nodes and wait it out through multiple negotiation cycles before moving on to the other nodes?  Or is this even possible?


Condor-users mailing list
To unsubscribe, send a message to condor-users-request@xxxxxxxxxxx with a
subject: Unsubscribe
You can also unsubscribe by visiting

The archives can be found at: