[ntp:questions] Remaining synced on an unsynchronised peer?
Brian Utterback
brian.utterback at sun.com
Tue Dec 1 17:45:29 UTC 2009
Dave Hart wrote:
> distrusted by downstream ntpds. Assuming no orphan-mode
> configuration, it is not clear to me if the stratum is supposed to go
> the maximum and the synchronization (leap bits) to 11, or if they are
> expected to remain at the last values with only the root dispersion to
> indicate the problem.
>
This has been discussed before. With ntpd, once an association gets
out of leap-11, it never goes back. That is not true of xntpd.
> I pick a nit about calling a ntpd unsynchronized as soon as its
> sources are severed. A "freewheeling" ntpd simply maintaining the
> last known frequency compensation is initially still a good source of
> time, decreasingly so over time. NTP 4 maintains an error budget
> along those lines, resulting in the root dispersion value that's part
> of every NTP response.
Which is exactly the reason an association never goes
"unsynchronized". I think the key milestones are:
1. Reach goes to 0.
2. Root dispersion goes to greater than 1 second.
3. Ref time goes to greater than one day.
As I recall, these milestones can trigger different behaviors, but I
don't remember just what they are.
More information about the questions
mailing list