[time-nuts] Divide by five ->Ensemble
tic-toc at bodosom.net
Sun Nov 9 18:06:49 EST 2014
On Sun, Nov 9, 2014 at 5:22 PM, Bob Camp <kb8tq at n1k.org> wrote:
> What is going on is that people are confusing the estimation process that
> is used by the selection process (which does look at a lot of stuff) and
> how that is described. They are then making the leap to the locking
> process, which is something else altogether.
Have you read the referenced documentation? It quite clearly describes
five major components:
- Clock Filter Algorithm
- Clock Select Algorithm
- Clock Cluster Algorithm
- Mitigation Rules and the prefer Keyword [viz. Clock Combining
- Clock Discipline Algorithm
Admittedly saying "Mitigation Rules and the prefer Keyword" rather than
saying Combining Algorithm could lead one astray but there is a step
between Cluster and Discipline. As noted earlier this bit: "designate one
of them as the *system peer*"
does not mean one clocks offset is used to discipline the client system
It's easy to see that this is what's going on by looking at the
> performance of an NTP implementation.
I'm not sure what you mean here but "root distance" degradation dictates
that no Stratum N+1 clock can be "better" than its upstream Stratum N
> In certain cases, with equally good / very good clocks, the selection
> process falls apart and the output actually degrades compared to the best
> clock in the group due to "best clock" selection changes.
Do you mean this?
"In practice, with fast LANs and modern computers, the correctness interval
can be quite small, especially when the candidates are multiple reference
clocks. In such cases the intersection interval might be empty, due to
insignificant differences in the reference clock offsets."
More information about the time-nuts