Auto-Adaptive scheduler and semaphore patch ( 2.2.14 ) ...

[David Lang]

On Tue, 25 Jan 2000, Davide Libenzi wrote:

> Hi David,
> 
> Tuesday, January 25, 2000 9:15 PM
> David Lang <dlang en diginsite.com> wrote :
> > Davide, your patch did a couple of things, one of which was to improve the
> > cache hit rate by re-ordering things. would it be possible to seperate
> > that portion of the patch out so it does not skew the results?
> 
> can You explain better ?
> I don't understand what You mean about reordering.
>

if your new code reorders stuff  to fit the cache better, but the old code
does not then your new code will do better in comparisen then if both sets
get teh optimization.
 
> 
> > for those who are saying that this doesn't match any real-world
> > situations, if this really does give a benifit aroung 4-5 processes in the
> > runqueue then it may be a valid selection for a 4 or 8 way SMP box. I
> > would not want to use on on a UP box or a 2 way SMP box, but if it becomes
> > a compile time option for large servers it may be reasonable.
> 
> It is not necessary, since for low workloads the new patch use the old
> scheduler.

if you dynamicly change from one to the other how can you benchmark the
two to see where the proper place to switch over is? Also the test that
you have in to use one or the other is pure overhead, for benchmarking it
would be better to have two kernels, one with each scheduler and no test
to choose between them.

As I was pointing out if the crossover point is low enough it may end up
being correct to use the new scheduler all the time for boxes with more
CPUs

David Lang



> 
> Davide.
> 
> --
> All this stuff is IMVHO
> 
> 
> 
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo en vger.rutgers.edu
> Please read the FAQ at http://www.tux.org/lkml/
> 


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo en vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/