[ntp:questions] Remaining synced on an unsynchronised peer?

David Woolley david at ex.djwhome.demon.invalid
Tue Dec 1 08:13:26 UTC 2009


Richard B. Gilbert wrote:
> unruh wrote:

>> do not know, that part of the problem is the level. The
>> machines will always have a level one higher than the system they are

systems, not system.  ntpd gets its time from a mix of that of the times 
  on all the systems listed with a * or + in ntpq peers.

>> getting their time from, and they do not like to get time from a system
>> at a higher level than they are. Thus the first one to get its time from

Some of the stratum sensitivity has been relaxed in version 4, but I 
think that version 3 will consider sources that have a stratum one or 
two worse than their own, or at least ones equal to their own.

A quick look at some of the version 4 code suggests that all strata 
except 16 contribute to the time, subject to not being falsetickers, and 
not exceeding the maximum number of sources.  The system peer is has the 
numerically lowest stratum, subject to:  their not being a valid prefer 
peer with a high numerical stratum and the previous system peer not 
being still valid, a higher numerical stratum, and having been at the 
lowest numerical stratum sufficiently recently (anti-clock hopping).

>> the other will up its level to one higher than the other, and the other
>> will then see the first as having too high a level(stratum) to sync to.

You are now talking about the selection of system peer, which is not the 
same as that of its time sources, although it will be one of them.

>> (or the two will walk each other up to the highest level of 15 and see
>> each other as unuseable sources)
>>
> No NTPD will take time from a Stratum greater than its own.  It's not a 
> matter of "not liking", a proper NTP client cannot take time from a 
> lower stratum (larger number).  Stratum One gets time from an atomic 

I'm not convinced that is true for version 3 (I don't know for version 
4).  Even if it is true....which looking at the code is not the case....

> clock.  Stratum Two gets time from Stratum One, Stratum Three gets time 
> from Two and so on.  In principle strata range from 1 to 15.  In 

Stratum three gets time from stratum two and stratum 1.  It is possible 
that there is a stratum 1 server that looks worse than its stratum 2 
system peer.

> practice anything claiming a stratum between 6 and 15 is probably bogus.
> 
>> I might be wrong in that analysis but that is my suspicion.
>>
> I disagree.  I too may be wrong but I don't think so.




More information about the questions mailing list