Too much pauzed Gamma-Ray Pulsar Search # 1.03 and Gravitational Wave S6GC search 1

Quote:

It's quite counter-intuitive as the newly started tasks often seem to be in much less deadline stress than the 'closer-to-deadline' suspended tasks. I think BOINC actually does know what it is doing despite the apparent paradox.

I agree with Gary's diagnosis, but not his prognosis.

Fred, your report reminds me of things I've seen when boincmgr thought I was in deadline trouble on a project. Sometimes it was right, and sometimes it was wrong because something had bumped up the completion time estimate way too high (such as suddenly thinking my host slow because of one fouled-up result, or thinking I was not going to use it at a decent rate in the near future because of a few day's vacation).

But the really, really strange thing is to see just what you described--task after task started up, usually NOT from the ones needed soonest, and then shortly paused to run yet another. I've seen this get above several dozen, which is pretty odd.

I, personally, have dealt with such episodes with direct intervention, even though I'm usually on the "don't fiddle" side of the endless argument. My preferred intervention in this case is to put the unstarted jobs on the project in trouble on suspend, so that boincmgr can't continue the mad dash through the list. Then I put all save a few of the soonest due on suspend as well, then exit boincmgr and restart, thus avoiding the memory and possibly other overhead of all that paused work.

Oh, yes, I suspend work fetch just in case boincmgr should be tempted to add to the chaos--probably don't need to, but I do.

Sadly, this approach requires some continuing followup attention during recovery, taking some tasks out of suspend enough in advance that it always has work available, still selecting stuff due soon, until the available work as currently estimated by boincmgr is little enough not to pose a threat.