We observe that on our grid, the Negotiator considers only 2000 jobs each cycle, where 2000 is the number of available slots. We have usually 4-5000 jobs queued, but roughly 1000 will be periodic in that they will want to rerun every 5-15 mins. This seems to have effect that only 1000 non-periodic jobs are considered, so we often see under utilisation. All jobs are submitted by the same user, and we're (still) onÂ htcondor 7.8 (linux).
Is the limit we're observing real, and is there a way to force the Negotiator to consider all the currently queued jobs in every cycle.