Re: [Ntp] Antw: Re: Antw: Re: Antw: [EXT] NTPv5 Loop Detection without Stratum

[Ulrich Windl <[e-mail]>]

>>> Miroslav Lichvar <[e-mail]> schrieb am 31.08.2022 um 10:39 in
Nachricht <Yw8eVn34M+Ip6ey6@localhost>:
> On Wed, Aug 31, 2022 at 09:52:01AM +0200, Ulrich Windl wrote:
>> I think that is not sufficient: It would be a loop if the server's clock
>> (actually the error estimate) gets worse again, and then the process 
> continues
>> (clock is getting more inaccurate over time). However it it's "smoothed
out"
>> it's not a loop IMHO.
> 
> The impact of the loop depends on how much weight the source has,
> polling intervals, number of servers in the loop, and other things.
> Even if the clocks can stay close to the true time, there is still a
> negative impact on stability.
> 
> When the conditions are right, the loop can amplify oscillations. In
> this example (which I have shown before) there are 4 ntpd servers
> using the same source and also synchronizing in a A‑>B‑>C‑>D‑>A loop.
> 
> https://fedorapeople.org/~mlichvar/ntpresponse/sync‑loop.png 
> 
> Do you blame the user for misconfiguration? Or would you prefer if the
> protocol enabled the servers to detect the loop?
> 
>> > In the graph theory, it's a loop in a directed graph.
>> 
>> The problem is the "line" (correpsonding to time): In NTP it's a
combination
>> of several sources (lines).
> 
> Which means the loops can be ignored?

Well, every phone call is a "loop": The microphone records what it hears from
the other end of the line (and sends it back, where the other end does the
same), too.
If the amplification is too high you'll have a never-ending loop of noise; if
it's low, it will vanish.
I think NTP should only forbid the first example of a loop.

> 
> ‑‑ 
> Miroslav Lichvar