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

Re: [HTCondor-users] [CondorLIGO] Different NEGOTIATOR_PRE_JOB_RANK for parallel and non-parallel universe jobs?

The NEGOTIATOR_PRE_JOB_RANK boolean expression evaluates in the context of each machine classad (including anything you publish in the machine ad from job ads of the jobs already running there, via STARTD_JOB_ATTRS), so you can simply reference the universe and give it a different rank.  Something like (in pseudocode):

NEGOTIATOR_PRE_JOB_RANK = (is_parallel * 10) + (is_not_parallel * 20) + other_stuff


Peter F. Couvares

> On Mar 3, 2015, at 9:12 AM, Steffen Grunewald <Steffen.Grunewald@xxxxxxxxxx> wrote:
> The default NEGOTIATOR_PRE_JOB_RANK setting, in a partitionable-slot
> setup, favours machines which have already been partially matched,
> by producing a lower rank for smaller CPU counts and memory sizes.
> Unfortunately, Parallel Universe jobs will get matched against
> such machines first as well, when it seems to be favourable to have
> a maximum of "locality", i.e. to use machines with high core counts
> first. I know that there is a penalty imposed by thermal throttling
> (which may be addressed by advertising less CPU cores) and - to me
> at least - seems smaller than the network overhead added by spreading
> a 100-rank MPI job over >50 machines.
> Is there a way to apply a different NEGOTIATOR_PRE_JOB_RANK expression
> to parallel universe and non-parallel universe jobs, respectively?
> And how can I have a look at such macro values across the pool?
> Thanks, S
> _______________________________________________
> Condorligo mailing list
> Condorligo@xxxxxxxxxx
> http://lists.aei.mpg.de/cgi-bin/mailman/listinfo/condorligo