I'm experimenting with Kernel Preemption

Dave Jones davej at redhat.com
Wed May 11 21:01:08 UTC 2005

On Wed, May 11, 2005 at 02:17:25PM -0300, Juan Carlos Castro y Castro wrote:
 > Not on Fedora's kernel sources, but with 2.6.11ac7. I based my config on 
 > /boot/config-2.6.11-1.14_FC3 and deat with the additional options with 
 > "make oldconfig". Then I browsed the configuration with "make 
 > menuconfig", just for fun.
 > I saw kernel preemption was turned off, so I turned on. Afterwards, I 
 > notice the system is noticeably faster. Bootup is faster. Shutdown is 
 > faster. The Red Hat manu on GNOME pops up WAY faster. OpenOffice.org 
 > loading is faster. I suspect other things are faster too, but I'd have 
 > to time them.

Every time this comes up, theres no concrete numbers. Just
'it feels faster'. Given the best that preempt can do is
lower the _average_ latency, rather than worse-case latencies
which many folks believe, I find it hard to believe it makes
a noticable difference. In a blind-test, given two kernels,
I'd bet on you not being able to 'feel' which one had
preempt enabled.

 > So my question is: why isn't preemption enabled in the FC3 packaged 
 > kernel? Does it conflict with something I haven't encountered yet? maybe 
 > some esoteric hardware combination? My hardware data is below.

It doesn't really buy anything worthwhile, and adds complexity and
opportunity for drivers to break.

 > Another thing: what crucial patch, if any, am I missing by using 
 > 2.6.11ac7 instead of the FC3 packaged kernel?

Exec-shield is probably the biggest feature, but there are a bunch
of other minor features (restricted /dev/mem, ipw wireless for eg).


More information about the users mailing list