Re: [SCHED] wrong priority calc - SIMPLE test case

[Mike Galbraith <efault@gmx.de>]

At 01:05 PM 1/14/2006 +1100, Con Kolivas wrote:
>On Saturday 14 January 2006 03:15, Mike Galbraith wrote:
>
> > Um... try irman2
> > now... pure evilness)
>
>Hrm I've been using staircase which is immune for so long I'd all but
>forgotten about this test case. Looking at your code I assume your changes
>should help this?

Yes.   How much very much depends on how strictly I try to enforce.  In my 
experimental tree, I have four stages of throttling: 1 threshold to begin 
trying to consume the difference between measured slice_avg and sleep_avg 
(kidd gloves), 2 to begin treating all new sleep as noninteractive (stern 
talking to), 3 to cut off new sleep entirely (you're grounded), and 4 is 
when to start using slice_avg instead of the out of balance sleep_avg for 
the priority calculation (um, bitch-slap?).  Levels 1 and 2 won't stop 
irman2, 3 will, and especially 4 will.

These are all /proc settings at the moment, so I can set set my starvation 
pain threshold from super duper desktop (all off) to just as fair as a 
running slice completion time average can possibly make it (all at 1ns 
differential), and anywhere in between.

         -Mike