[ntp:questions] NTP on LAN

David Woolley david at ex.djwhome.demon.co.uk.invalid
Mon Sep 29 06:56:07 UTC 2008

David L. Mills wrote:
> I'm not sure which configuration you are considering; the documentation 
> clearly shows, even in the diagrams, that there can be more than one 

Steve, rather than I, was proposing a configuration in which there was 
one orphan mode server and one normal client.  I have no problem with 
the use of orphan mode with multiple servers at the same orphan level.

> orphan server and more than one orphan client. The expressed intent is 
> that, should all sources of time be lost, that onlyu all orphan clients 
> select the same orphan server. It's as simple as that.
> Orphan mode was motivated by certain misguided configurations that tried 
> to engineer failover configurations using engineered local clock stratum 
> assignments. These are dangerous, fragile and will form loops unless 

Steve was talking about engineering a failover precedence arrangement by 
  engineering the orphan strata to all be different.

> specifically and carefully done. Usually under some failure scenario or 
> other a cycle forms between two local clocks and in the best 
> Bellman-Ford tradition, they count to infinity.
> Orphan mode was specifically designed for the most common case where a 
> LAN has multiple Internet servers and where one or more of them can 
> become disconnected. The (single) orphan server becoms the source of 
> last resort.

I'm not sure if you mean single in a static or dynamic s sense.  The 
documentation indicates benefits when it is a dynamic sense, but doesn't 
seem to indicate any real difference from the local clock solution when 
there is never more than one candidate.

More information about the questions mailing list