Kernel timers

[John DeDourek]

Callum Lerwick wrote:

> On Thu, 2006-06-15 at 14:07 +0100, David Woodhouse wrote:
> 
>>On Thu, 2006-06-15 at 02:32 -0500, Callum Lerwick wrote:
>>
>>>On Wed, 2006-06-14 at 23:33 +0100, David Woodhouse wrote:
>>>
>>>>Tickless operation. We need to abandon the timer tick.
>>>
>>>Interesting, what would this mean for low latency operation?
>>
>>It would mean that the resistance to switching to 1000HZ gets massively
>>reduced.
>>
>>Basically, the current implementation stops the timer tick when the
>>machine is _idle_. This means that we're not waking the machine up every
>>1ms and wasting power, and that was the main reason we didn't want
>>HZ=1000 in the past.
> 
> 
> Perfect! I know such patches have been floating around for years now,
> (And apparently some platforms have been using them already for a while)
> What's the current holdup? :)
> 

What are the implications on accurate conditioning of the system time by
ntp?  The kernel on my FC4 machine is showing some bad indications
kernel-2.6.16-1.2115_FC4) relative to accurate conditioning.  Basically,
the time correction as plotted from the ntp time daemon oscillates
+1 and -1 ms. about zero, starting a day or so after a reboot.  I haven't
had time to track this down.  But it is awkward for people trying to
accurate system and network measurements.

My question to the development team is whether ntp clock conditioning is
a consideration or not when evaluating changes to the kernel time keeping
functions.