[ntp:questions] NTP topology design question

Richard B. Gilbert rgilbert88 at comcast.net
Thu Jan 17 16:08:27 UTC 2008

Erik Soosalu wrote:
> I remember seeing something about this before, but I can no longer find it, so I'm going to ask here.
> I've got a Garmin GPS17HVS driving a HP thin-client and I seem to be getting good results from it.  This setup is sitting in my head office and is currently working very well as I can tell.
> Now I'm looking at a next steps - is it wise to setup another ntp server as a peer?  I'm thinking of setting up another server at a geographically remote location (~50ms over the private wan) and having that peer with the existing box.  All critical equipment would be setup with both NTP boxes as their servers.
> Thoughts?
> Thanks in advance,
> Erik
> _________________________________________________________________

Two servers is the worst possible configuration.  If the two disagree, 
which one do you believe?

The fifty millisecond delay is not good either!  Is that the round-trip 
delay or is it one-way?

Four servers are the minimum for a robust configuration; any one of the 
four can fail (no response or incorrect time) and the other three will 
out vote it.  Five servers allow two failures and seven servers allow 
three failures.

I'd let the two run independently and rely on "things equal to the same 
thing are equal to each other".  The best that peering could do would be 
to guarantee that they are within 1/2 the round trip delay of each other.

More information about the questions mailing list