Standard set of terms for precision, accuracy, related concepts.

[Achim Gratz]

Eric S. Raymond writes:
> Achim, you and Gary *both* get to write glossaries covering terms like
> precision, accuracy, drift, and related stuff.  Give it your best
> shot(s).  If, after a reasonable period of time, I have a glossary
> only from one of you, tha person wins and the glossary gets blessed
> and added to the official documentation.

That is a bit of a can of worms as you have already seen, both in the
original thread and the answers here.  Just the two terms you mention
are used in different ways for different things and so far we haven't
even determined whether we can or want to unify on one set of
definitions throughout all of NTP or maybe keep domain-specific meanings
where appropriate.  Another line along which these terms split is
whether they are applied to continous or discrete quantities, whether
you are talking about a single number or a statistical moment and
whether the thing you are talking about is a stochastic variable, a
physical quantity or something you do a calculation with.

For NTP, I'd think the only physical continous quantity of interest is
time, but it only ever gets processed as a quantized numerical value.
NTP als cannot directly measure the time, instead it approximates it by
various means, in particular in the form of time differences.  That
approximation requires both statistical inference and numerical
calculations.  So there are at least three domains that need a glossary:
representation of absolute and delta time (both conceptual and in
implementation), the statistical inference and the implementation of
these operations as numerical algorithms.

Many of the terms used for ntpd are actually referring to internal
algorithmic variables of the control loop rather than an estimate of
some measureable quantity.  The frequency offset for instance does
describe the deviation of the system clock from the ideal frequency only
when both the derivative of the frequency offset and the time offset are
both zero.  In all other cases it's a mixture of the (not measured)
offset of the clock frequency and an additional offset introduced by the
FLL/PLL that tries to keep the time offset as close to zero as
possible.  Similarly the time offset isn't the actual offset, but a
measurement corrupted by multiple noise sources.


Regards,
Achim.
-- 
+<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+

DIY Stuff:
http://Synth.Stromeko.net/DIY.html