Thanks for your reply, also to Kevin. This makes sense and explains the behaviour I'm seeing. I now know that I have to make some other choices in my setup, but that's OK.
Best regards and all the best for 2026 (for everyone), Maurice On Wed, Dec 31, 2025 at 09:27:51PM +0300, Kihaguru Gathura wrote: >Hi Maurice > > >This is expected behavior with OpenNTPd and not a problem with your >hardware. > >With only a local sensor, ntpd can control the clock very tightly so that >microsecond offsets are normal but once you add remote servers, their >network delay and asymmetry start to influence the clock control as well. > >Even if the sensor remains selected in ntpctl, the other peers will still >affect the filtering and frequency correction, which typically results in >millisecond-level variation. > >Weights don’t fully isolate the sensor, and seeing it selected does not >mean the other sources have zero impact. > >If you want to keep microsecond accuracy, the usual approach is to run the >machine as a pure stratum-1 server with only the sensor configured and get >redundancy by running multiple such boxes. > >However, If you need fallback to network time in the same daemon, then >OpenNTPd is simply not designed for that use case. > >Regards, > >Kihaguru Njenga Gathura > > >On Wed, 31 Dec 2025, 17:28 Maurice Janssen, <[email protected]> wrote: > >> Hi, >> >> I have a couple of small machines (Soekris 6501 and 5501) that I use as >> NTP server in the NTP pool. >> When I use a sensor as the only source of time (Garmin 18lvc on the serial >> port or Meinberg PZF180PEX DCF77 receiver card), the offset is quite good: >> nearly allways below 20 us, most of the time below 5 us and sometimes even >> less than 1 us. >> But when I add another server (mainly as backup in case the sensor doen't >> work), the offset drifts away to 1 or 2 ms. I tried with 1 sensor and >> 1 server, with 1 sensor and multiple servers, I tried adding weight 5 or 10 >> to the sensor, but this doesn't seem to have any effect. In all cases ntpd >> stays synchronised to the sensor (with poll=15s), but after an hour or so >> after starting ntpd the offset starts to increase. And from that moment, >> the >> offset wanders between roughly -2 and +2 ms. >> It almost seems as if ntpd is synchronised with one of the servers (with >> much higher polling interval), however ntpctl still shows that it is >> synchronised to the sensor (albeit with the wandering offset as described >> above). >> >> I would prefer to have the microsecond offset _and_ some servers as backup, >> but can't get it working like that. >> >> Am I missing something? Or is this a bug in ntpd? >> >> Thanks for any help. >> >> >>
