[ntp:questions] Re: NTP and reference-source evaluation?

David Woolley david at djwhome.demon.co.uk
Sun May 22 17:59:10 UTC 2005

In article <42909A94.2F9E1442 at en-ninguna-parte.com>,
ben at en-ninguna-parte.com wrote:

> Does the current v4 and previous v3 NTP algorithms evaluate a server or
> peer's source of time, at all?

The quality of the time is evalatued by using a combination of the
stratum, which is why LCL clocks should always be fudged to a high
stratum, root delay (network propagation delay uncertainty) and
root dispersion (and estimation in the cumulative drift of measured
times due clock resolution and clock frequency drift).  Stratum has
a very high priority.

> That is, when a local client or a peer gets time from a remote server or
> peer, can or does the local client or peer additionally evaluate the

It prefers low stratums, low delays and low dispersions.  However, this
breaks for LCL clocks as they tend to report root delay and root dispersion
as zero when the true error could be arbitrarily large.  Fortunately the 
high priority of stratume means that LCL clocks will be effectively 
ignored in the presence of valid reference clocks.

In my view, LCL clocks should report large root dispersions and root
delays.  Unfortunately accurate values would take the root distance
outside the maximum allowed and cause the source to be ignored entirely.

More information about the questions mailing list