On Mon, Mar 28, 2011 at 10:10:29AM -0300, Luis Claudio R. Goncalves wrote: > On Mon, Mar 28, 2011 at 09:40:25PM +0900, Minchan Kim wrote: > | On Mon, Mar 28, 2011 at 02:28:27PM +0200, Peter Zijlstra wrote: > | > On Mon, 2011-03-28 at 21:21 +0900, Minchan Kim wrote: > | > > Hi Peter, > | > > > | > > On Mon, Mar 28, 2011 at 11:51:36AM +0200, Peter Zijlstra wrote: > | > > > On Fri, 2011-03-25 at 00:27 +0900, Minchan Kim wrote: > | > > > > > | > > > > At that time, I thought that routine is meaningless in non-RT scheduler. > | > > > > So I Cced Peter but don't get the answer. > | > > > > I just want to confirm it. > | > > > > | > > > Probably lost somewhere in the mess that is my inbox :/, what is the > | > > > full question? > | > > > | > > The question is we had a routine which change rt.time_slice with HZ to > | > > accelarate task exit. But when we applied 93b43fa5508, we found it isn't effective > | > > any more about normal task. So we removed it. Is it right? > | > > | > rt.time_slice is only relevant to SCHED_RR, since you seem to use > | > SCHED_FIFO (which runs for as long as the task is runnable), its > | > completely irrelevant. > | > > | > > And Kosaki is about to revert 93b43fa5508 to find out the problem of this thread > | > > and Luis said he has a another solution to replace 93b43fa5508. > | > > If rt.time_slice handleing is effective, we should restore it until Luis's patch > | > > will be merged. > | > > | > Right, so only SCHED_RR is affected by time_slice, it will be > | > decremented on tick (so anything that avoids ticks will also avoid the > | > decrement) and once it reaches 0 the task will be queued at the tail of > | > its static priority and reset the slice. If there is no other task on > | > that same priority we'll again schedule that task. > | > > | > In short, don't use SCHED_RR and don't worry about time_slice. > | > | There was meaningless code in there. I guess it was in there from CFS. > | Thanks for the explanation, Peter. > > Yes, it was CFS related: I think it wasn't related CFS but O(1). I guess as we changed O(1) with CFS, the fault was remained. -- Kind regards, Minchan Kim -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxxx For more info on Linux MM, see: http://www.linux-mm.org/ . Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/ Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>