[time-nuts] NIST time services
phk at phk.freebsd.dk
Tue Mar 25 03:13:25 EDT 2014
In message <E1WSEMC-000Dz8-34 at stenn.ntp.org>, Harlan Stenn writes:
>> I'm actually not certain that it helps, even if you document it.
>> It's sort of an "administrative" distance and it unfairly penalizes
>> any GNSS in favour of terrestial if you calibrate it according to the
>> original intent...
>I'm game to come up with a better plan. Original intent is good, and
>follow-on improvements are even better.
Internal to NTPns I used a different scheme to pick refclock, each
refclock got one vote for each independent source it had.
This means that GPS almost always wins, since it has one vote for
each sat not thrown out of the solution, whereas for instance
DCF77 or WWVB would only get one vote, there being only one transmitter.
Propagating something like that directly out of the S1 server is
probably not sound, we don't want everybody to use the furthest
server in TTL, just because it has a 24 channel GPS receiver.
Likewise, NIST and USNO may have significantly more faith in their
coax cable than the rest of us should have in over-the-air signals.
So all in all, I'm not really sure what we can do automatically with
the root dispersion at S1 level.
At lower levels I think it works more or less as is should.
Poul-Henning Kamp | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG | TCP/IP since RFC 956
FreeBSD committer | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.
More information about the time-nuts