Re: [HTCondor-devel] Priorities between accounting groups

[Jon Thomas <jthomas@xxxxxxxxxx>]

On Mon, 2013-01-28 at 10:09 -0800, Igor Sfiligoi wrote:
> But this does not really manage fair share between groups...
> e.g. if for one day G1 uses all the slots because there are no jobs from G2,
> on day two, when G2 gets back, G1 will still get 33% of all the available slots, right?

There is no historical weighting at the group level.

> Also, what happens if I have
> G1=.25, G2=.25, G3=.5
> and there are only G1 and G3 jobs in the queue?
> Will G3 get twice the "opportunistic" slots compared to G1? Or is there no such guarantee?
> 
With GROUP_ACCEPT_SURPLUS=true, the slots will be consumed
G1=.25/.75=.33 and G3=.50/.75=.67 .

> Thanks,
>    Igor
> 
> On 01/14/2013 11:59 AM, Jon Thomas wrote:
> > On Fri, 2013-01-11 at 09:06 -0800, Igor Sfiligoi wrote:
> >> Dear HTCondor team.
> >>
> >> I am puzzled on how priorities between accounting groups work.
> >> Some time ago I have asked around (don't remember who I talked to, sorry), I was told that if I set the dynamic quotas to >>1.0, then I effectively get priorities between them this way.
> >> (e.g. G1=2.0&G2=4.0, users in G2 get 2/3 of the pool on a busy day, bust still can get all the resources if G1 is not there)
> >>
> >> Looking in the 7.8 manual, I see instead that "each of the subgroups will have their dynamic group quota scaled"!
> >> http://research.cs.wisc.edu/htcondor/manual/v7.8/3_4User_Priorities.html#26789
> >>
> >> Is this a recent change?
> >> Dan tells me there has been a lot of work in this area.
> >>
> >> Please let me know.
> >> If the semantics has indeed changed, how can I get the desired semantics back?
> >
> > Looks like what you are after is G1=.33, G2=.67 and
> > GROUP_ACCEPT_SURPLUS=true.
> >
> >>
> >> Thanks,
> >>     Igor
> >> _______________________________________________
> >> HTCondor-devel mailing list
> >> HTCondor-devel@xxxxxxxxxxx
> >> https://lists.cs.wisc.edu/mailman/listinfo/htcondor-devel
> >
> >
>