Re: radeon graphic speed regression 2.4 -> 2.6 test7 benh
On Thu, 2003-10-30 at 14:04, P Oscar Boykin wrote:
> On Thu, Oct 30, 2003 at 09:11:16AM +1100, Benjamin Herrenschmidt wrote:
> > I doubt that even a working CONFIG_PREEMPT would make any good
> > difference. Kernel preemption is just a crappy feature to please
> > lame slashdotters
> >
> > Ben.
>
> Are you joking? I have heard that it is a pretty good feature for
> audio recording and other applications that need as close to real-time
> performance as possible (I guess live video processing, recording,
> etc...)
>
> Why do you say it is a crappy feature?
Not the place to argue about it here, but I'd rather continue fixing
the kernel latency problems if any (we did fix a lot recently) that
just preempting when in kernel space. Also, that force all of the SMP
stuffs on UP kernels, thus slightly increasing overall kernel overhead
on non-SMP setups.
Ben.
Reply to: