Re: [Condor-users] starving jobs with high RequestCpus

[Matthew Farrellee <matt@xxxxxxxxxx>]

On 01/22/2011 03:20 AM, Erik Aronesty wrote:
> Has anyone noticed that jobs that request, say, 6 cpus tend to languish
> for far more than 6 times longer than jobs which need 1 cpu.
>
> Does condor ever say to itself... OK, this job is old, the user has a
> very low effective priority... and I'm going to wait until i get 6 cpus free
>
> Or do the 1 cpu jobs keep jumping in forever... preventing the 6-cpu job
> from running?
>
> - Erik

Condor only gathers slots like that in the Parallel Universe.

Most approaches to avoid the potential starvation you're talking about 
are variations on

   https://condor-wiki.cs.wisc.edu/index.cgi/wiki?p=WholeMachineSlots

and are addressed purely in policy, as opposed to Negotiator or 
Scheduler code.

Best,


matt