[ntp:questions] Re: NTP to sync clocks on an Isolated Local Network (UTC not relevant)

Tom Smith smith at cag.zko.hp.com
Fri Sep 1 13:38:41 UTC 2006


duncan.perrett at elekta.com wrote:
> Nero,
> 
> Thanks for the prompt reply.
> 
> My application will be safety critical so if any microprocessor fails
> then the whole system has to move to a safe state/reset/shut down.
> Therefore I can't see any advantage in using more than one clock as the
> reference clock.  I don't mind depending on a single machine!
> 
> Would the synchronisation be improved?  Remember I'm not worried about
> the "stability".
> 
> Assuming I did follow your advice, how is that achieved?  Do I need
> peer commands or something?
> 
> Do my conf files look OK if I wanted to stay with my original set-up?
> 

Given that there is no good setup for NTP in the absence of any
precision clocks, your original proposal is about the best you
can do. Configuring all systems to use their local clocks is
precisely the wrong thing to do. They will form little cliques
among themselves, each with a different time. If any redundancy
were required, configure at most 3 such servers with strata
incrementing by 2 (e.g., 8, 10, 12) and configure the rest as
simple clients. I would configure drift files everywhere except
on the single server in your originally proposed broadcast
configuration.

Note also that many, if not most, users who claim to be concerned
only about having the same time and not about having the correct
time on a group of systems are shocked to learn a few weeks later
that their clocks are running fast or slow and have drifted by
several minutes or several hours off "real" time. Clocks running at
the wrong speed is a direct, and possibly more important, consequence
of not having a precision clock anywhere.

-Tom




More information about the questions mailing list