[LinuxPPS] Recommendation / Re: Stats & architectures

[Bernhard Schiffner]

Am Donnerstag, 13. März 2008 15:23 schrieb Remco den Besten:
...
> I believe that the Ulrich Windl patches included nanosecond precision
> while LinuxPPS is at the microsecond level.
>
> Is it possible to implement this nanosecond precision in conjunction
> with LinuxPPS?
>
Nearly everything is possible and nearly nothing makes sense.

Please don't mix accuracy and  resolution as a first step.

A TSC is the finest grained source (<ns) of timing information. But this 
source is questionable (processor-specific). Others (HPET) do a little bit 
less and so on.
(BTW: What does your /sys/.. entry say about your clocksource?)

And it takes a long way until an interrupt is handled and compared with 
timing information. Bus, SuperIO, context switching, cacheing  add a "big" 
amount of time until something happens or can happen again.
(Use 30µs as a very first guess.)

And keep in mind: there is no proof of accuracy within pure ntp. It forms 
only "receiver(s)" . It takes other programing and hardware to confirm 
something down to the close-µs level.
(Don't speak about ns.)

Conclusion: Don't worry to much.

Bernhard