Hi!
I have just cooked up a patch which mixes Mike Galbraith's excellent
monotonic clock O(1) scheduler changes with another patch I think that
came from Ingo Molnar and some scheduler parameter tweaks. This patch is
against 2.5.72-mm3, but applies cleanly on top of 2.5.73.
For me, it gives impressive interactive behavior. With it applied, I can
no longer make XMMS skips sound, moving windows on an X session is
perfectly smooth, even when moving them fastly enough for a very long
time.
Thanks for listening!
On Mon, 23 Jun 2003 21:41, Felipe Alfaro Solana wrote:
> Hi!
>
> I have just cooked up a patch which mixes Mike Galbraith's excellent
> monotonic clock O(1) scheduler changes with another patch I think that
> came from Ingo Molnar and some scheduler parameter tweaks. This patch is
> against 2.5.72-mm3, but applies cleanly on top of 2.5.73.
>
> For me, it gives impressive interactive behavior. With it applied, I can
> no longer make XMMS skips sound, moving windows on an X session is
> perfectly smooth, even when moving them fastly enough for a very long
> time.
Hi Felipe.
For those who aren't familiar, you've utilised the secret desktop weapon:
+ if (!(p->time_slice % MIN_TIMESLICE) &&
This is not how Ingo intended it. This is my desktop bastardising of the
patch. It was originally about 50ms (timeslice granularity). This changes it
to 10ms which means all running tasks round robin every 10ms - this is what I
use in -ck and is great for a desktop but most probably of detriment
elsewhere. Having said that, it does nice things to desktops :-)
Con
On Mon, 2003-06-23 at 16:15, Con Kolivas wrote:
> For those who aren't familiar, you've utilised the secret desktop weapon:
>
> + if (!(p->time_slice % MIN_TIMESLICE) &&
>
> This is not how Ingo intended it. This is my desktop bastardising of the
> patch. It was originally about 50ms (timeslice granularity). This changes it
> to 10ms which means all running tasks round robin every 10ms - this is what I
> use in -ck and is great for a desktop but most probably of detriment
> elsewhere. Having said that, it does nice things to desktops :-)
I lost the track to Ingo's patch sometime ago, so I borrowed it from
your latest patchset ;-) It does really nice things on desktops. It
brings 2.5 to a new life on my 700Mhz laptop.
What impact would have increasing MIN_TIMESLICE from 10 to, let's say,
50?
Thanks!
On Monday 23 June 2003 13:41, Felipe Alfaro Solana wrote:
Hi Felipe,
> I have just cooked up a patch which mixes Mike Galbraith's excellent
> monotonic clock O(1) scheduler changes with another patch I think that
> came from Ingo Molnar and some scheduler parameter tweaks. This patch is
> against 2.5.72-mm3, but applies cleanly on top of 2.5.73.
> For me, it gives impressive interactive behavior. With it applied, I can
> no longer make XMMS skips sound, moving windows on an X session is
> perfectly smooth, even when moving them fastly enough for a very long
> time.
I am using your patch ontop of 2.5.73-mm1. Only two words: perfectly great!
> Thanks for listening!
thanks for the effort and the patch!
ciao, Marc
On Tue, 24 Jun 2003 04:54, Felipe Alfaro Solana wrote:
> On Mon, 2003-06-23 at 16:15, Con Kolivas wrote:
> > For those who aren't familiar, you've utilised the secret desktop weapon:
> >
> > + if (!(p->time_slice % MIN_TIMESLICE) &&
> >
> > This is not how Ingo intended it. This is my desktop bastardising of the
> > patch. It was originally about 50ms (timeslice granularity). This changes
> > it to 10ms which means all running tasks round robin every 10ms - this is
> > what I use in -ck and is great for a desktop but most probably of
> > detriment elsewhere. Having said that, it does nice things to desktops
> > :-)
>
> I lost the track to Ingo's patch sometime ago, so I borrowed it from
> your latest patchset ;-) It does really nice things on desktops. It
> brings 2.5 to a new life on my 700Mhz laptop.
>
> What impact would have increasing MIN_TIMESLICE from 10 to, let's say,
> 50?
It stops being helpful :) 50 was the default Ingo used but this was far too
long for say X to be waiting just to get to move the mouse again - this is
why the mouse movement is so much less jerky with this set to 10.
Con