[time-nuts] Win XP and NIST time

[Magnus Danielson]

Hi Chris,

On 03/26/2013 04:30 PM, Chris Albertson wrote:
> I would think that the best cure for persistent server abuse (abuse
> that continues even after a KoD is sent) would be to send back a bogus
> time with a huge random error of many thousands of hours.  A normal
> NTP client will notice the error and stop sending requests and the
> simple ntp clients in Windows and home routers will annoy their users
> so much they would get disconnected.
>
> A lot of people don't understand how NTP works and think sending many
> packets will work better.   It is simple to understand if you think
> about how to set a real mechanical wall clock.    The goal is to set
> the "fast/slow" lever to the "correct" position.  So you set you clock
> to match a standard clock then wait and after waiting you look at the
> offset from the standard.  The LONGER you wait the better.  NTP's goal
> is to adjust the rate of your clock.  At first it waits a short time
> then as your clock's rate gets closer to correct it waits longer and
> longer.  The longer times are actually a good sign.  BUT somehow
> people think forcing a one second or one minute interval is better.

Yes and no. As long as you are on the white frequency modulation noise 
of the local source, increasing the time-constant time is good, until 
you reach the flicker frequency modulation noise. It's the cut-off 
between them which forms the limit in that balance.

There is another aspect, in that the amount of polling rate controls how 
well you are able to filter out the noise which steers how precise time 
become on the link. This is in fact an orthogonal property and goal to 
that of the frequency tracking.

The polling rate and time constant of the filter remains two different 
things, aiming to solve two different problems with two different ideals.

Cheers,
Magnus