[ntp:questions] Re: where should we update sys_rootdispersion?

David L. Mills mills at udel.edu
Tue Sep 9 03:04:27 UTC 2003


Indeed if you batter ntpd long and hard enough you can cause the
frequency to clamp at the extreme and it may take quite a while to
settle down, especially if symmetric modes are involved. The system is
actually a network of chaotic oscillators with nonlinear feedback loops
coupled by a nonuniform sampled data stream. Professional driver needed;
don't try this at home.

Patience. Give it a couple hours.


Dongling Duan wrote:
> Thanks David for your explanation!
> I have a script which has more than ten testcases using the topology I
> describe in previous email. I test all kinds of configuration and
> combination, server, peer, broadcast, with/without authentication etc.
> For each test, I reset clock to a random time on all box to see whether
> we are synced or not after some time. Some times offset could be very
> very big. Once it is synched, we check whether each peer is reached via
> peer->reach.
> Could somehow ntpd internal state fall into some state (sys_jitter is
> very big) because of all the configure/unconfigure, reset clock etc? So
> the ntpd just need more time to be sane?
> -dongling

More information about the questions mailing list