LKM Timemark Driver

Gary E. Miller gem at
Tue Aug 28 19:57:38 UTC 2018

Yo Achim!

On Tue, 28 Aug 2018 21:43:03 +0200
Achim Gratz via devel <devel at> wrote:

> Gary E. Miller via devel writes:
> >> That doesn't really matter, it is still about three orders of
> >> magnitude improvement over what we get today for the error that
> >> matters.  
> >
> > Ah no.  Lukas says 200 ns.  I can easily get under 1,000 ns.
> > Sometimes down to 400 ns with care.  
> So do I, in fact I run between 140ns to 240ns of "average jitter"
> pretty consistently across five machines (three different rasPi and
> two TinkerBoard).

I was talking Standard Deviation, a bit better measure of the jiter.

Can you post your ntpvix URL?

> However, those numbers are essentially
> useless, as I'm quite certain that the true residual offset is at
> least 10…15µs and plausibly into the 30µs range.

Yes, studies have shown the consumer GPS can be stable in the 10 to
20 ns range, but the offset from "true" GPS can be much worse.

> I could tune out
> some of the offset by comparing the average deviation as observed in
> a three-cornered-hat configuration over long observation times, but
> that would still leave a residual offset.  Getting that residual
> offset down to roughly the same figures, only in the nanosecond range
> instead of microseconds is what I was talking about.

That will need a multi-frequency GPS receiver.  Even a multi-channel
GPS receiver has difficutly getting under 10 ns before post proecssing.
The GPS signal itself is just not that accurate.

> > That is 2x, maybe 5x, certinaly not three
> > orders of magnitude.  You'll have 3 orders of magnitude when you
> > get to 1 ns jitter.  No consumer GPS can do that.  
> Stop deluding yourself with meaningless numbers.

They are all we have.  If you can create some more meaningful numbers
then please send patches to ntpviz.

For now we can use our numbers as arbitrary goodness numbers to compare
two configurations.

>  Even if loopstat
> started to tell you such low figures it would simply be a lie.

Agreed, I also never believed loopstats.

> To get
> below 1ns you need a clock source that has significantly better ADEV
> than 1e-9 @ tau=1s when free running.

Yup, got that.  I think my Rb is 10e-14.  At least that is what the
calibration papers said two months ago.

Gary E. Miller Rellim 109 NW Wilmington Ave., Suite E, Bend, OR 97703
	gem at  Tel:+1 541 382 8588

	    Veritas liberabit vos. -- Quid est veritas?
    "If you can’t measure it, you can’t improve it." - Lord Kelvin
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 851 bytes
Desc: OpenPGP digital signature
URL: <>

More information about the devel mailing list