Subject: Re: Configurable scheduler quantum?
To: Matthias Buelow <mkb@mukappabeta.de>
From: Thorbjorn Jemander <thoan@ifm.liu.se>
List: netbsd-users
Date: 05/03/2001 16:28:25
Matthias Buelow wrote:
> 
> Thorbjorn Jemander writes:
> 
> >10 ms by setting the rrticks to hz/100. If I want values
> >lower than that, say 5 or 3 ms I have to up the HZ value,
> >but how does that affect the rest of the system? When
> 
> I wonder why you would want to do such a thing at all, generally
> one would like to have as large quantums as possible without (human-)
> noticable delays in context switching.  This has emperically been
> determined to be about 1/10s (of course originally in larger
> timesharing environments, where many users were using the computer
> through a teletype or tube terminal, mostly typing at the keyboard
> but I think Vahalia (can't check now) reports in his book that the
> time quantum is also remarkable accurate in modern days (graphical,
> mostly single-user environments) towards user expectations).  Do
> you have any special use for your system in mind that would benefit
> noticably from smaller quantums?  I'd be interested in what that
> might be.

This is a long story for which I can't go into detail right now
(preparing my thesis defense), but it has to do with creating a
desktop system with the "illusion" of being fast and responsive,
while actually sacrificing the system performance. Reducing the
time quantum is one part of it. In order to make it effective
I also need preemted user-land threads (which is worked on, I believe)
and some other things. It may turn out to be impossible under
NetBSD, but I am looking around to see which stumbling blocks there
are. I know that X11 is one of them, another might be that it's
not considered "the unix way". At this point, this is not too
serious.



/Thorbjörn Jemander