On Apr 28, 2009, at 12:00 PM, [email protected] wrote: > The problem with ignoring possible triggers is that there is no way of > knowing when the next trigger will occur. It could be in a second, it > could be hours, there is no way of knowing in advance. > > I know I proposed this before. > > Normally run in RR. > If there is a deadline issue detected by the rr_sim, schedule some > of the > tasks based on EDF. However, do NOT necessarily preempt immediately. > If there is also a deadline problem detect by EDF, THEN we have to > preempt > immediately. > > Note that the rr_sim has a bit of a hair trigger. EDF does not. > > As to the checks taking longer on faster machines. Won't the speed > of the > machine speed up the test as well?
Yes it will as a matter of absolute clock time ... BUt, we have a cascade effect here. More processors do more work, creating more triggers, spending more and more time obscessing over the schedule to no great advantage. So, though the increased speed of the processors allows more work, and the absolute time, and maybe even the proportional time stays the same, the question is still when do we accept that we have hit a point of diminished returns. _______________________________________________ boinc_dev mailing list [email protected] http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev To unsubscribe, visit the above URL and (near bottom of page) enter your email address.
