Kernel scheduling algorithm and X.Org performance

[Jim Gettys]

Yes, certainly the Sun work is interesting/applicable (don't let the
date on it scare you).  But the drivers are a fundamental problem that
needs fixing: right now, X looks like (is) a CPU pig, and gets scheduled
accordingly.

Another approach is one we've discussed with Ted T'so and Steven Tweedie
and others: the X server might trade its scheduling quanta back and
forth with clients so that clients (at least in the local case) would
get full use of their scheduling quanta.  Anyone feel like doing Linux
patches?  The problem is making sure that scheduling credits are
conserved.
			- Jim




On Wed, 2005-08-31 at 10:29 -0700, Alan Coopersmith wrote:
> Dmitry Shatrov wrote:
> > I would like to talk about X-based desktop responsiveness and how close
> > it turns out to be related to the kernel's process scheduling algorithm.
> 
> Sun did some work on this years ago, and ended up giving the X server the
> ability to boost the kernel scheduling priority of the window manager and
> the process with focus.
> 
> The research that went into this was presented at the 1993 Usenix - the paper
> is available at:
> http://www.usenix.org/publications/library/proceedings/cinci93/full_papers/evans.txt
> 
> I've been promising for a while to make the source available for X.Org and
> finally got it up today at  https://bugs.freedesktop.org/show_bug.cgi?id=2192 .
> 
> This is simply a dump of the code currently in Solaris, and not something ready
> to go into Xorg - before that happens, it would be useful to figure out if this
> is helpful on other platforms/modern hardware, and if so, how should we make the
> code more portable/generic (as noted in the bugzilla comments).
>