Tony,
Watch out for the sampling theorem. That has to do with the
minimum
frequency for sampling time-varying signals with specified
bandwidth.
What set this discussion off was the notion on the part of
some folks
that accuracy could be defined in some arbitrary manner. Not
so, and my
carefully crafted error budget is the counter. I would take
serious bets
even now that few have actually seen and considered the
draft
specification. I have received no comments on the draft of
any kind.
Dave
anthony.flavin bt.com wrote:
> This is an interesting discussion, but it should be
well understood by
> most people.
>
> We are dealing with nothing more that sampling theorem.
The fact that the
> quantity we are sampling is time is irrelevant. Terms
such as precision
> and resolution are well understood here. As are error
(noise).
>
> If these terms have been re-defined within NTP, then
that's fine and those
> definitions should hold.
>
> So what was the original question that was trying to be
answered? Because
> I have lost track....
>
> Regards,
>
>
> Tony Flavin
> Core Networks Engineering Manager
>
> e-mail: anthony.flavinbt.com
> tel: +44(0) 1473 609570
> mobile: +44(0) 7801 759596
> fax: +44(0) 1908 862752
>
> Orion 5, pp11 , Adastral Park, Martlesham, Ipswich IP5
3RE, UK
>
> P.s "Big Ben" is the bell not the clock:-
>
> The Great Clock
> Hands: Minute (copper) Length 4.3m (14 ft) Weight
101.6kg (2 cwt)
> Hour (gunmetal) Length 2.7m (9 ft) Weight 304.8kg (6
cwt)
> Pendulum: Total Length 4.4m (14 ft 5 in)
> Length of Roman numerals: 61cm (2 ft)
> Minute squares: 30.5cm (1 ft)
> Number of panes of glass 312
> in each clockface
> The Bells
> Big Ben (the Great Bell) Weight 13.8 tonnes 13 tons 10
cwt 99 lb
> Note E: Hammer Weight 203.2kg 4 cwt
> Quarter Bells
> 1 Note G sharp Weight 1 ton 1 cwt 23 lb 1.07 tonnes
> 2 Note F sharp Weight 1 ton 5 cwt 30 lb 1.28 tonnes
> 3 Note E Weight 1 ton 13 cwt 69 lb 1.71 tonnes
> 4 Note B Weight 3 tons 10 cwt 69 lb 3.59 tonnes
> Time between Strikes:
> From start of chime to 12th strike = 95 seconds
> Big Ben: From 1st strike to 12th strike 54 seconds, 5
seconds between
> strikes
>
>
>
> -----Original Message-----
> From: ntpwg-bounces+anthony.flavin=bt.comlists.ntp.isc.org
> [mailto:ntpwg-bounces+anthony.flavin=bt.comlists.ntp.isc.org] On Behalf
> Of Kurt Roeckx
> Sent: 20 February 2007 23:14
> To: David L. Mills
> Cc: ntpwgntp.isc.org
> Subject: Re: [ntpwg] Timing Definitions
>
>
> Hi Dave,
>
> On Tue, Feb 20, 2007 at 12:56:05AM +0000, David L.
Mills wrote:
>
>> Stand next to Westminister Abbey and set your watch
to Big Ben. Big
>> Ben
>> clockkeepers keep the clock in tune by listening to
the BBC and putting
>> coinage on the pendulum. Presumably, the
"accuracy" of Big Ben depends
>> on the residual frequency error and the interval
between coinage
>> updates. You set your watch to Big Ben and inherit
its characteristics
>> and some of your own. How "accurate" is
your watch?
>
>
> Since I have no idea about what the accuracy or
precision is of Big Ben,
> it's going to be rather hard to say anything about it.
And I think that's
> the point we're both trying to make.
>
> Anyway, there are 2 ways to synchronize your clock to
Big Ben. One is
> looking at it and reading the time. This is as far as I
know with a
> resolution of 1 minute, so a precision of atleast 1
minute. But the clock
> might be set with an accuracy in the order of a few
seconds. The other is
> listening to it's bells which is going to have a higher
precision but a
> different accuracy.
>
> To say anything useful about any of those 2 ways, you
need to do
> statistics by comparing it to a clock that has a small
enough error.
>
> If I'm going to set my watch by any of those 2 methods,
what matters is
> which way has the smallest total error. The accuracy of
my watch is going
> to depend on that, and not how accurate the clock is or
how precise I can
> "read" it.
>
> You can go and reduce the error caused by the precision
by taking more
> samples, but you can't compensate for the accuracy
error.
>
> Anyway, the definition of precision and accuracy I use
might be a little
> bit different than what you're used to when talking
about NTP. If you take
> some samples and compare it to a reference, the
accuracy would be the
> average difference between the samples and the
reference, the precision
> would be the standard error.
>
> What you use in NTP for the precision is the highest of
the resolution and
> the time it takes to read the clock. This is a
reasonable estimate for
> how precise you can measure the local system clock.
It's not a good
> estimate for how precise a remote ntp daemon can come
and read your system
> clock since it doesn't only depend on your own system
now. It's this
> "remote" precision that is used to
synchronize the clock. And I think you
> try to estimate that by the jitter.
>
>
> Kurt
>
> _______________________________________________
> ntpwg mailing list
> ntpwglists.ntp.isc.org http
s://lists.ntp.isc.org/mailman/listinfo/ntpwg
_______________________________________________
ntpwg mailing list
ntpwglists.ntp.isc.org
http
s://lists.ntp.isc.org/mailman/listinfo/ntpwg
|
