Re: Priority Inversion in Scheduling

[Mike Galbraith <efault@gmx.de>]

At 07:35 AM 9/10/2003, Mike Fedyk wrote:
>On Wed, Sep 10, 2003 at 06:42:10AM +0200, Mike Galbraith wrote:
> > At 02:23 AM 9/10/2003, Nick Piggin wrote:
> > >Hi John,
> > >Your mechanism is basically "backboost". Its how you get X to keep a
> > >high piroirity, but quite unpredictable. Giving a boost to a process
> > >holding a semaphore is an interesting idea, but it doesn't address the
> > >X problem.
> >
> > FWIW, I tried the hardware usage bonus thing, and it does cure the X
> > inversion problem (yeah,  it's a pretty cheezy way to do it).  It also
> > cures xmms skips if you can't get to the top without hw usage.  I also
> > tried a cpu limited backboost from/to tasks associated with hardware, and
> > it hasn't run amok... yet ;-)
>
>Against which scheduler, and when are you going to post the patch?

Against stock test-4, but I'm not going to post it.  It's just an 
experiment to verify that there is another simple way to defeat the X 
inversion problem (while retaining active list requeue).  Also, backboost 
is a tricky little bugger, and I thought I'd let Nick know that I had some 
success with this heavily restricted form.  (global backboost can be down 
right evil)

If anyone having inversion or concurrency troubles wants to give it a try 
for grins, they can drop me a line.  My tree tends to morph a lot though, 
depending on what aspect of scheduling I'm tinkering with at the time.  It 
currently does well at defeating known starvation issues, but I don't like 
it's priority distribution much (and it's not destined for inclusion, and 
it's pretty darn ugly, and I'll likely break it all to pieces again soon, 
and...;).

         -Mike