Re: [chrony-dev] PPSAPI: kernel consumer

[Håkan Johansson <f96hajo@xxxxxxxxxxx>]

> > I've tested chrony's PPS driver and I'm really shocked with its
> > synchronization quality. It is a HUGE improvement over ntpd. It was
> > hard to believe that it can sync clock with a precision of 1-2us but
> > it is true. However, my new kernel consumer seems to be able to keep the
> > same precision, but it can sync faster (in few seconds).
>
> I think chrony could do that too if the kernel interface allowed a
> tight loop.
>
> This could be an extension to adjtimex which would allow us to know
> either remaining singleshot adjustment (ideally in nanoseconds) or
> when exactly did the last change in frequency happen.

I'd agree that a user-space application should be able to synchronise just 
as well as the kernel, provided the interfaces for both controlling the 
clock _and_ doing accurate measurements are available.

I've recently (for slightly other reasons) measured the in-kernel time 
consumed by doing getnstimeofday() (several in a row), and for most 
machines (2 GHz+) it took a whopping ~ 500 ns.  I.e. one cannot know to 
better than 0.5 us what the time was, if one e.g. try to sandwich 
something else with two such calls.  The best machine was actually a 
slower one (~1.05 GHz) that had about 60 ns between return values for two 
back-to-back getnstimeofday().  I suppose we are down to PCI bus latencies 
and so on.

Synchronisation would mean both to determine the offset, and the gain to 
keep the clock correct futurewise?  With the gain being the harder part 
--- if measurement precision is 1 us, then over a 1 s time-span, only 1 
ppm accuracy is obtained.

Cheers,
Håkan