Perhaps my statement was ambiguous, here's a next try. Although in the meanwhile we found the culprit and it's NOT ntp, I still consider it worth to post the solution as others might experience similar problems. Sorry for the noise.
Our Setup: Stand-alone (Ubuntu) Linux desktop computer, GPS/PPS clock connected to serial port, kernel and ntp configured appropriately. We use this box to do accurate delay measurements. I.e., only the local clock must be synchronized to UTC, there's no timeserver functionality provided to others. The ntp.conf file has only one single server entry for the NMEA/PPS clock (server 127.127.20.0 and corresponding fudge line). Half an hour or an hour after system start, ntpq -p reports the offset to be in the order of 1-5 microseconds - perfect for our measurements. But whenever we changed the measurement setup (i.e., unplugged the Ethernet cable from the network interface and plugged it in again) the ntpq -p-reported offset suddenly increased to several hundreds of microseconds and took again at least half an hour to settle to the required value (below 10 microseconds). This delayed all of our measurements. We suspected ntp to listen to the ifups and ifdowns and restart timekeeping on such events. To make it short, the culprit was Ubuntu's default NetworkManager configuration. The file /etc/network/if-up.d/ntpdate explicitly (re)starts ntp on any network interface that comes up (either by plugging in the connector or by issuing an ifup command). As simplest solution you can remove the rights on the file (sudo chmod 000 ntpdate) such that your ntp will no longer lose the time on network interface status changes. hope it helps and sorry for the noise, br Joachim Am 12.06.2015 um 12:26 schrieb Marco Marongiu: > On 12/06/15 11:54, Joachim Fabini wrote: >> The straight-forward solution that you propose was the one that I tried >> first. Unfortunately it does not work. > > Then I guess I don't understand what you're after exactly. I'll re-read > your message entirely and check what I misunderstood. Sorry for the noise. > > -- M > _______________________________________________ > questions mailing list > questions@lists.ntp.org > http://lists.ntp.org/listinfo/questions > _______________________________________________ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions