Re: [Ntp] Temperature Compensation for NTP?

[Kurt Roeckx <[e-mail]>]

On Tue, Dec 08, 2020 at 01:50:52PM -0800, Harlan Stenn wrote:
> On 12/8/2020 1:44 PM, Hal Murray wrote:
> > 
> > [e-mail] said:
> >> Given we have better understanding now of the thermal stability of solid
> >> state (crystal) oscillators, why not include that into the algorithm if we’re
> >> going to flesh it out specifically? 
> > 
> > I think the algorithm(s) should be split out to a separate document, if any.
> 
> In which case somebody will only implement the protocol and do whatever
> they want with the algorithms, and clients will connect to this server
> and there will be a significant chance that because of the different
> algorithms, the behavior will be different and there will be train wrecks.

I think we should not say how it should be implemented, but explain what
is expected of what is put in all the fields. For instance, it
should explain how to calculate the errors, and what you
optionally do to minimize the errors.

> Different timing algorithms will cause much bigger problems.
> 
> NTP is all about a defined impulse response.

I think defining NTP as a impulse response is the wrong approach.
It's an implementation detail. If other systems use different
algorithms, it should only have an effect on the uncertainty. If
it causes big problems, it's an implementation problem.


Kurt