NTP Kernel Discipline
This page shows a couple of plots based on the NTP statistics files from two FreeBSD systems being fed by PPS sources (one an HP Z3801A GPSDO, the other a Spectracom 8164 WWVB receiver). The plot covers one day. At the beginning of the plot, NTP was running without kernel timekeeping discipline, but at about 1300 UTC NTP was restarted with the kernel discipline enabled (using the "fudge 3 1" flag on the PPS source).The second set of plots are of about two hours with the kernel timekeeping discipline enabled, showing the much smaller apparent offset and jitter. But are the systems really performing any better?
