[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Condor-users] Order of selection of machines in condor pool
- Date: Wed, 12 Oct 2005 12:57:57 -0500
- From: Erik Paulson <epaulson@xxxxxxxxxxx>
- Subject: Re: [Condor-users] Order of selection of machines in condor pool
On Wed, Oct 12, 2005 at 11:25:45AM -0500, Steven Timm wrote:
> I recently added 40 fast nodes to my pool of what had been just 14
> slow nodes. The 40 fast nodes seem to be working OK. However,
> when the pool is mostly idle, it seems like it is very often the
> slow nodes that match first. for example, there are 16 jobs
> running at the moment and all but one of them have matched on the slow
> nodes that were in the pool at first.
> So what is the scheduling algorithm that is being used? As far
> as I know any node in the pool would meet the requirements of
> these jobs equally well. I verified by submitting 100 jobs that
> all the nodes are in fact able to accept and run jobs. It's obviously
> not a round-robin scheduler. What would make the bias?
When there are no preferences at all, Condor gives out matches in no
defined order. (And we'll never put anything by default in, because we
have no idea what's best. Should we prefer fast machines to slow machines?
Should we prefer "dedicated" to "desktop" machines? More RAM versus
less RAM? Too many options...)
Instead, we punted to the user. Check out NEGOTIATOR_PRE_JOB_RANK:
It lets you sort the list of machines that a job matches with.
If you want to steer jobs to fast machines, you can set it to be
NEGOTIATOR_PRE_JOB_RANK = TARGET.Mips
> Steve Timm
> Steven C. Timm, Ph.D (630) 840-8525 timm@xxxxxxxx http://home.fnal.gov/~timm/
> Fermilab Computing Div/Core Support Services Dept./Scientific Computing Section
> Assistant Group Leader, Farms and Clustered Systems Group
> Lead of Computing Farms Team
> Condor-users mailing list