LKM Timemark Driver

Gary E. Miller gem at rellim.com
Tue Aug 28 21:28:30 UTC 2018


Yo Achim!

On Tue, 28 Aug 2018 22:56:37 +0200
Achim Gratz via devel <devel at ntpsec.org> wrote:

> Gary E. Miller via devel writes:
> >> 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.  
> 
> As you should know, the jitter distribution is not normal,

Yes, which is why ntpviz provides skewness and kurtosis so you 
can tell how the distribution differs from a standard distribution.

ntpviz also supplies a number of other bits about the data set:
mean, average, min, max, etc.

> so strictly
> speaking the standard deviation you talk about doesn't exist.

Nope.  Standard Deviation applies to ALL data sets.

https://en.wikipedia.org/wiki/Standard_deviation

    "In statistics, the standard deviation (SD, also represented by the
    Greek letter sigma σ or the Latin letter s) is a measure that is
    used to quantify the amount of variation or dispersion of a set of
    data values."

People are taught in college how Standard Deviation looks for a 
"normal distrubution', so they assume that one denotes the other.

>  You can
> however obtain a similar measure of distribution width that converges
> to the standard deviation for normal distributions under fairly loose
> conditions.

Which is useless for us, because, as you rightly noted, we have nothing
at all that looks like a normal distribution.

The skewness and kurtosis was added to ntpviz to make this clear.

> > Can you post your ntpvix URL?  
> 
> No, it's an internal network and I won't change that.

Not a requirement to post ntpviz data.  Many people upload their
ntpd from their stratum one to another server to do the computations.

> Besides, as you
> well know, I don't run ntpviz.

Thus my request that you do.

> I'm pretty sure I posted my plots
> before however, I can check later this week to point you at the
> archives.

Which would be OLD data.  I'm looking for LIVE data.

> > 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.  
> 
> Again, I wasn't talking about that number.  The GPS module's internal
> time representation, for all we know, is the ground truth towards
> we're supposed to coverge.

No, we are trying to converge on UTC(USNO).

> > They are all we have.  If you can create some more meaningful
> > numbers then please send patches to ntpviz.  
> 
> You need some independent measurement to get there.  The ntpviz plots
> will never show an independent view.

Yup, which is why we are working to get the PPS-out patch into the
kernel.  That coupled with my TAPR-TICC and Rb will give us that
external confirmation.

So hopefully 'never' is this year.

> > For now we can use our numbers as arbitrary goodness numbers to
> > compare two configurations.  
> 
> Yes, but you keep making statements that go far beyond that confidence
> interval.

We'll have to disagree on that.  My data is open to all to see, in real
time.  I'm open to suggestions on how to improve it more.

> > Yup, got that.  I think my Rb is 10e-14.  At least that is what the
> > calibration papers said two months ago.  
> 
> Read it again.  You're not getting that performance at tau=1s, not
> even close.

I did dig it out, my mistake.  I'm "only" getting 1.5e-11 at tau=1s.  I
guess I'll have to settle for that.

> In fact you're unlikely to get that performance even for
> larger tau unless you discipline it via GPS (in which case you get
> there maybe at around 2…4 weeks).

At which time you are measuring the GPS, not the Rb.  And a time scale
that is useless for measuring the short term jitter we are discussing.

And, to answer the next question.  Yes, I am discipling my Rb against a
GPS.  The additional accuracy is very small.  Maybe another 10x.

RGDS
GARY
---------------------------------------------------------------------------
Gary E. Miller Rellim 109 NW Wilmington Ave., Suite E, Bend, OR 97703
	gem at rellim.com  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: <https://lists.ntpsec.org/pipermail/devel/attachments/20180828/b43500a0/attachment-0001.bin>


More information about the devel mailing list