Fedora Core 2 wishlists

[Nicolas Mailhot]

Le lun 08/12/2003 à 20:31, Felipe Alfaro Solana a écrit :
> On Mon, 2003-12-08 at 19:47, Ori Pessach wrote:
> 
> > | No, the pre-emptible kernel feature doesn't seem to be quite ready
> > | yet.  However, the 2.6 kernel also has the lowlatency patchset applied,
> > | and that gives pretty much all of the benefit.  Clark Williams wrote
> > | a good paper on this a while ago.
> > 
> > What are the issues? (a pointer to the Clark Williams paper would be
> > fine as an answer). I saw a dramatic improvement in scheduling latency
> > when I recompiled the 2.6 test 11 kernel to enable the pre-emptible
> > kernel. When compared to the RPM available from
> > http://people.redhat.com/arjanv/2.5/RPMS.kernel/, scheduling latency was
> > down by an order of magnitude, if I remember correctly. If that kernel
> > RPM doesn't contain the lowlatency pathces, that's understandable, but
> > otherwise I'm not sure what's going on.
> 
> There are some strange, not-easily-reproducible bugs that seems to get
> triggered when preemptible is enabled.

However Linus' opinion seems to be this is not preempt code that is
buggy, but preempt makes it a lot easier to hit *existing* races.
So preempt should certainly go in rawhide and maybe even be available as
a debug kernel later.

> > It would be nice not to have to recompile the kernel to get this
> > functionality.
> 
> Currently, it's impossible to do that, since pre-emption support can't
> be enabled as a module. Enabling preemption on-the-fly is not an easy
> task so, for the time being, you'll need to recompile.

And since preempt touches core code it probably can not even be
modularised.

Cheers,

-- 
Nicolas Mailhot
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: Ceci est une partie de message num?riquement sign?e.
URL: <http://listman.redhat.com/archives/fedora-devel-list/attachments/20031209/e6b7f69a/attachment.sig>