Messages in this thread | | | Date | Sat, 26 Jun 2004 11:05:32 +1000 | From | kernel@kolivas ... | Subject | Re: [PATCH] Staircase scheduler v7.4 |
| |
Quoting Willy Tarreau <willy@w.ods.org>:
> Hi Con, > > although I was one of those who complained a lot about the 2.6 scheduler, > and still don't use it because of its sluggishness under X11, I'm impressed > by your work here. I've tried the good old test which was *very* sluggish > on a vanilla 2.6 :
Well I can take the blame for that since I tuned it. I tried very hard to do it without changing the overall design :\
> > # for i in $(seq 1 20); do xterm -e sh -c "while :; do locate /;done" & done > > It opens 20 xterms constantly listing my slocate database (vmstat shows > no I/O). > > Under vanilla 2.6 (up to 2.6.4 at least), some of these xterms would freeze > for up to about 10 seconds IIRC during redrawing, with incomplete lines, > etc... > This still happens with your patch and /p/s/k/interactive=0, but to a lesser > extent it seems. But it does not happen anymore with interactive=1, hence > the > progress ! > > However, you warned us that the nice parameter was very sensible. Indeed, > it *is* ! When my window manager (ctwm, very light) is at 0, just like the > script above, the windows appear slowly and irregularly on the screen, > but this takes no more than 15s, during which windows get no title, then > suddenly they get everything right. If I renice the WM at +1, I see no > more than 5 windows on the screen with no decoration at all, and then I > cannot even change the focus to another one anymore. Then, as soon as I > change the WM's nice value to -1, suddenly all remaining windows appear > with their title. The same is true if I start the script with the WM at > -1 initially. It's just as if the nice value was directly used as the > priority in a queue. > > Oh and BTW, this is an SMP box (dual athlon). > > Well, I see there is some very good progress ! Please keep up the good > work !
Thanks! I'll try my best.
Con
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |