Subject: Re: Scheduler project status and further discussion
To: Mindaugas <firstname.lastname@example.org>
From: David Laight <email@example.com>
Date: 01/18/2007 20:46:33
On Tue, Jan 16, 2007 at 07:39:06PM +0200, Mindaugas wrote:
> David Laight <firstname.lastname@example.org> wrote:
> > That is unlikely to be useful unless:
> > - you assign a cpu to the scheduler
> > - one scheduler is higher priority than any other.
> Well, it *IS* useful, or maybe I misunderstood something.
> Different schedulers could use different data (locks, counters, bitmaps, etc),
> different runqueue structures. Why not to allow to allocate and manage this
> data for scheduler's module? IMHO, this is a good structural abstraction.
All true, but at some point you have to some code that actually runs the
process. This needs (AFAICT) some sort of global queue of runnable processes
by priority - common to all schedulers.
> > > As an example, think of Mindaugas' Linux-like scheduler.
> > Is that the one that refuses to allow high priority processes to pre-empt
> > low priority ones for large fractions of a second ? :-)
> Heh.. which was that?
Behaviour I suspect the linux 2.6 default scheduler of having...
David Laight: email@example.com