Preemptible kernel in Fedora Core 2

Adrian Likins alikins at redhat.com
Wed Dec 10 20:34:03 UTC 2003


On Tue, Dec 09, 2003 at 03:04:52PM -0500, Michael K. Johnson wrote:
> On Tue, Dec 09, 2003 at 10:06:03AM -0700, Ori Pessach wrote:
> > comes from the lowlatency patches. The 2.6 kernels I tried (from the
> > RPMs I mentioned earlier) had really bad scheduling latency when the
> > pre-emptible kernel feature wasn't compiled in. They behaved just like
> > the 2.4 kernel without lowlatency.
> 
> Existing lowlatency patches were included.  That doesn't mean that no
> new latency points have crept in since.  "lowlatency patches" means
> fixing high latency points by explicitly yeilding; pre-emption is
> implicit yields.
> 
> I would suggest that using kernel profiling to track down additional
> latency points, since you are clearly hitting latency points now,
> would be a useful thing to do.
> 

	i've seen some improvements from using one of the arjan
kernels to useing arjan kernel+preempt while doing audio stuff.
It's hard to quantify, but ll+preemt doesn't cause overruns with
jackd, while the same settings with just ll sometimes causes them.

	That said 2.6+ll is still way way way better than the
stock rh2.4 for that kind of thing. 

	I'd say at least turn on preempt for rawhide, if
for no other reason than to find those mysterious bugs
no one can put a finger on. 

Adrian 





More information about the fedora-devel-list mailing list