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

[Miroslav Lichvar <[e-mail]>]

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?

-- 
Miroslav Lichvar