F7 redux, and the road to F8.

[Bill Rugolsky Jr.]

On Fri, Jun 08, 2007 at 01:24:05PM -0400, Dave Jones wrote:
> On Fri, Jun 08, 2007 at 05:50:50PM +0200, dragoran wrote:
> 
>  > this would mean that we will might end up having cfs as the scheduler 
>  > and tickless  x86_64.
>  > I mostly using x86_64 ... where there any major problems (exept the dell 
>  > one) related to tickless kernels in the F7 cycle?
> 
> Too early to say really. There were a number of oddball bugs
> that still aren't really understood that could be related,
> but on the whole it hasn't been /that/ bad considering the
> amount of code that changed.
 
I've generally been including Thomas's (pre-dynticks) hrtimers patch in
my custom kernels since 2.6.15, but after upgrading lots of boxes to FC6,
we've been running stock Fedora kernels (FC6 and FC7t*) on lots of boxes.
While there have been minor problems on x86, on x86_64 we experienced
severe NTP timekeeping regressions (including losing sync) due mostly to
various hardware latency problems (SATA chipsets problems, SMI, etc.).
The x86_64 hrtimers patchset works around the problem, and everything
keep times to within a millisecond.

So I hope to see the x86_64 tickless patches upstream ASAP.

Regards,

	Bill