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

[Con Kolivas <kernel@kolivas.org>]

On Tuesday 10 January 2006 02:52, Mike Galbraith wrote:
> At 12:11 PM 1/9/2006 +0100, Mike Galbraith wrote:
> >Care to try an experiment?...
>
> Oops.  I guess I should send one that's not mixed p1 and p0.  Sorry about
> that :-/
>
> Anyway, if anyone wants to see a functional demonstration, just try
> this.  Remove the TASK_NONINTERACTIVE in fs/pipe.c in both the stock kernel
> and this modified one so Davide Libenzi's excellent sleep pattern exploit
> (irman2) can work [1], and do the below all at the same time ...
>
> make -j4 bzImage
> irman2
> thud 3
>
> With the stock kernel, I got bored after a half an hour, and stopped the
> kernel build.  It had produced 40 .o files.  The modified kernel finished
> in 20 minutes vs the 8 minutes it took to produce the same 504 .o files if
> not under load.
>
>          -Mike
>
> 1.  it just so happens that Davide wrote irman2 using pipes... he could
> have done something else.  if anyone doesn't think this is a fair test,
> just use Paolo's much simpler exploit instead.  the result will be about
> the same.

Want to try with a few other schedulers using plugsched?

Con