Here is an updated version of the O1int patch designed to improve
interactivity.
This change addresses the difficulty of new tasks in heavy load being
recognised as interactive by decreasing the amount of time considered in the
interactivity equation, but dropping that decrease exponentially till it gets
to the MAX_SLEEP_AVG.
This should improve the startup time of new apps in heavy load and lessen
audio stalls when loads are high _and_ then the audio app is started.
Please test and comment.
Con
> Here is an updated version of the O1int patch designed to improve
> interactivity.
>
> This change addresses the difficulty of new tasks in heavy load being
> recognised as interactive by decreasing the amount of time considered in the
> interactivity equation, but dropping that decrease exponentially till it gets
> to the MAX_SLEEP_AVG.
>
> This should improve the startup time of new apps in heavy load and lessen
> audio stalls when loads are high _and_ then the audio app is started.
>
> Please test and comment.
>
Don't try this on a P4 with Hyper Threading enabled 8)
When doing a make -j6 even, mouse/keyboard/window switching/whatever
takes ages to respond. Worked fine without (even with a make -j20),
just tried this for fun.
Regards,
--
Martin Schlemmer
On Fri, 2003-06-27 at 05:26, Con Kolivas wrote:
> Here is an updated version of the O1int patch designed to improve
> interactivity.
>
> This change addresses the difficulty of new tasks in heavy load being
> recognised as interactive by decreasing the amount of time considered in the
> interactivity equation, but dropping that decrease exponentially till it gets
> to the MAX_SLEEP_AVG.
>
> This should improve the startup time of new apps in heavy load and lessen
> audio stalls when loads are high _and_ then the audio app is started.
>
> Please test and comment.
>
> Con
Well, I've just been following this thread and I choose to make a small
patch against your O1 int to export the scheduler tuning knobs via
sysctl. This patch will create /proc/sys/kernel/sched_tuning and expose
all the knobs, sou you guys don't have to recompile and reboot just to
change, say, max_sleep_avg. Don't know if anybody will find it useful,
but at least it doesn't hurts :-)
It should apply cleanly against plain 2.5.73, altough I didn't tested.
wrt your patch, I must admit I didn't feel any difference from vanilla
2.5.73, but this feeling isn't much scientific 8). I will do some
serious tests tomorrow if I find any time.
Roberto