Hi,
Am 23.11.2011 um 12:45 schrieb Lasse Knudsen:
> I have an interesting problem between two of my user groups they both
> share a queue, and i dont want to split the queue.
>
> What happens is that some jobs from group_A cause some jobs from
> group_B to fail, they both use a specfic independent consumable, so i
> was thinking that it might be possible using recource quotas to
> prevent jobs from group_B to execute on machines where there is
> already jobs running from group_A.
besides using two queue, there is a solution with two RQS and one queue:
- groupa and groupb are consumable complexes
- both are attached to each exechost with the number of slots per machine (in
this example 10)
- users of group A will request the complex of group B and vice versa
- two RQS (again for 10 slots per exechost):
{
name GroupA
description Limit reuti
enabled TRUE
limit users reuti hosts {*} to slots=10+$groupa*10-100
}
{
name GroupB
description Limit soft
enabled TRUE
limit users soft hosts {*} to slots=10+$groupb*10-100
}
Group A users submit (or by an JSV): qsub -l groupb=1 job.sh
As soon as one job of group A is running, the slot limit for group B becomes
zero or even negative. Other group A jobs can still start.
Cheeky user could use this to reserve an exechost for a group though (keep
always one job running there...)
-- Reuti
> I know the best solution was to have all jobs behave at all times, but
> this is not always possible.
>
> br
> Lasse
> _______________________________________________
> users mailing list
> [email protected]
> https://gridengine.org/mailman/listinfo/users
_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users
