[HTCondor-users] Parallel Universe and NEGOTIATOR_PRE_JOB_RANKing?

[Steffen Grunewald <Steffen.Grunewald@xxxxxxxxxx>]

(I haven't seen my yesterday's question on the list yet, so this
is also a test.)

With 8.2.0, the default NEGOTIATOR_PRE_JOB_RANK expression had been
changed (to something quite familiar), but this still doesn't seem
to result in best locality - perhaps since slots are "collected"
over time, until the requested number of machines are available?

Looking at the "machines" file for one of the currently running jobs,
(*) it's a quite unsorted collection of host names, and those show up
from 1 to $nslots times.

Apparently, not all nodes are occupied, though, which indicates
that the grade of randomness must be smaller than it looks at a 
first glance - is there a description how ParUniv matching is being
done, in particular if resources are tight?

(*) It would be nice if there was a consistent way to get this 
information at a central place, without having to find the node
running rank-0 (`condor_q -run`), then search the local execute/dir_*'s
there. Am I missing something obvious?

Thanks,
 Steffen