[time-nuts] Divide by five

"Björn Gabrielsson" bg at lysator.liu.se
Sun Nov 9 17:37:11 EST 2014

>>NTP does not pick the best clock.   NTP finds
>>the subset of clocks that track each other.
> NTP does indeed find the best clock from the subset of clocks
> which pass its sanity check, and then it uses only that one.
> There are several problems with that, and as we speak I'm developing
> a new algorithm which at least so far, gives much superior performance.
> You can read my musings about this here:
> 	http://phk.freebsd.dk/time/20141107.html
> My goal is to release the new NTP client before X-mas.
> --
> Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20


Looking forward to testing the new NTP implementations!

Many years ago I was bitten by a setup where ntpd did not exclude a

A normal PC (P2-450MHz) running linux (2.4.20ish-kernel with
ppskit-patches) was set up with 4 or 5 good internet ntp-servers as
reference. Two GPS-receivers were attached locally and configured as
refclocks. One Trimble with external event input (driver29) and a Rockwell
Jupiter with 1pps (hardpps) (driver31).

By mistake I disconnected the Jupiter antenna cable. After about day (day
and night, 24h), the clock had drifted ca 180ms. Even though all clocks
but the Jupiter was agreeing within 1ms of true time. The clock followed
the 1PPS of the drifting Jupiter oscillator, never minding the other

This could be since the jupiter 1pps, was configured as hardpps using the
kerneldriver, instead of letting ntpd using the pps in softpps(?)-mode.

Nevertheless I was a bit surprised that the computer continued to serve
time claiming to be a S1-server, even if both local refclock and a number
of external S1-servers said time was false. I dont know if recent ntpd and
operating system versions still have this problem.

Poul-Henning, for the updated server with local refclock, have you
elaborated over

1) supporting multiple 1PPS-sources on the same server?
2) supporting a stable frequency source? (say a ocxo or rb, that are
divided down to 1pps, 10pps or 1pp(minute), but the pulse is only stable
not aligned to a utc-second transition, and have a small frequency



More information about the time-nuts mailing list