I think the issue is that with 3 clocks, if 1 fails, you are left in a situation where NTP can't tell which remaining clock to trust.

I'd also be interested to know precisely where the servers are physically connected. While NTP is really very good at dealing with latency, it can't deal with asymmetry in the forward and return paths (it assumes one way delay = 0.5xRTT). I'd be curious to see how many people have symmetric routes to the servers.

In any case, I think it's a great move guys, and should be applauded. I have recently been playing with NTP and determining how accurately NTP over symmetric links matches GPS time and was pleasantly surprised by how close we could get. (Short story - a few tens of microseconds)

Cheers - N

On 3 September 2010 16:51, Jay Daley <jay@nzrs.net.nz> wrote:

On 3/09/2010, at 4:24 PM, Nathan Ward wrote:

> I note that there are three servers - I am surprised that there is not a fourth to allow for redundancy. Is that a cost decision?

I've always understood that three is enough for redundancy.  The client can tell when one goes awry by comparing with the other two.  Admittedly if one goes down then the client can't tell that but we can pursue that logic ad infinitum no matter how many servers there are.

Do you see it differently?

And yes, cost was a notable factor.

cheers
Jay

>
> Looks like a good project anyway! It'll be good to have some time servers that are formally publicly available and maintained - I never have been sure of the state of the ihug/massey/waikato/etc. NTP servers that I've configured boxes to use in the past.
>
> --
> Nathan Ward
> _______________________________________________
> NZNOG mailing list
> NZNOG@list.waikato.ac.nz
> http://list.waikato.ac.nz/mailman/listinfo/nznog