Kernel Preempt?

[Arjan van de Ven]

On Fri, 2005-07-08 at 01:32 -0400, Brian C. Huffman wrote:
> I see that preemption is disabled in the Fedora Core kernels.  What's
> the reason for this?  I've searched the lists and haven't seen any
> definitive responses although the possibilities range from problems w/
> SMP to difficult to reproduce issues.  
> 
> I'm running a Toshiba Laptop and I'm currently in the process of
> tweaking everything I can from a performance point of view (since
> laptops are inherently less responsive than desktops) and I'd like to
> know if 1) enabling this is dangerous 

from a distro pov preempt traditionally has been frowned upon because it
has/had a history of hitting obscure bugs. Which one person won't care
about but from a distro pov that's a problem

> and 2) will I see a perceptible
> latency reduction thus an increase in user experience?

no. preempt doesnt' actually reduce latency THAT much. And the fedora
kernels do have voluntary preempt enabled, which does help *worst case*
latency.

The reality is that even without the any form of preempt, the 2.6 kernel
gets latencies on average well below 1ms. Now worst case is a different
story (eg if you really really stress the box), but normally you don't
do that to laptops. (and to be fair, from an interactive pov you have
bigger problems if you do, you'll find that you are disk bound due to
paging in that case which latency benchmarks don't care about but
desktop experience does).

1ms is well below what a human can perceive so from a "user feel" point
of view it all doesn't really matter.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
Url : http://lists.fedoraproject.org/pipermail/test/attachments/20050708/5bd7d5fd/attachment.bin