On May 21, 2009, at 5:43 AM, john.mcl...@sybase.com wrote: > Paul: > > I agree that there is a problem. I disagree with your analysis on > what the > fix was. > > The problem is that BOINC pre-empts too easily. In some cases (not > all > cases though) the task switch to a high priority task could wait > until the > next normal task switch. > > You want to just make the test less frequent without fixing the > underlying > cause of the problem. I want the underlying problem fixed. > > A person entering an issue in TRAC should stick to the reproducible > facts > and should normally shy away from analysis.
You see, therein lies the rub, because preempting too easily is a separate issue. Not only do we preempt too easily, but the test is not limited to seeing if we have reason to preempt on checkpoint. The 5 times "enforce" was run in one second is exactly the problem I have been trying to corral with the no floor issue. I think that there are other issues that are related to the rate at which we run enforce. The current difficulty is that there are still a number of bugs in the way we schedule tasks to be run. Most of them are currently in the area of "enforce", there are a couple in the area of RR Sim and the rules therein. Maybe my analysis is/was wrong, it has been before, it will be again ... then again, so has yours. It is one thing to disagree and dismiss out of hand, it is vastly different to prove another wrong. You did the former, not the latter... And I too want the underlying problems fixed. And sometimes to fix the tire on the race car you actually have to slow it down to work on it... There has been a long history of behavior in the development of BOINC, if you are not one in a very small select group of like minded individuals your idea, your bug report, your whatever, will be summarily dismissed with a very weak rationale ... But, being a lone voice in the wilderness is not something I am unfamiliar with, and though I know that UCB developers, and now you, will simply dismiss anything I write or say ... I do it for *MY* honor because it is my duty to the BOINC Community to try to represent them ... I know I don't do it well ... but I do it as best as I can ... _______________________________________________ boinc_dev mailing list boinc_dev@ssl.berkeley.edu http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev To unsubscribe, visit the above URL and (near bottom of page) enter your email address.
