On Wed, Sep 13, 2023 at 15:52:26 +0200, Miroslav Lichvar wrote: > On Wed, Sep 13, 2023 at 09:28:34AM -0400, Josef 'Jeff' Sipek wrote: > > However, I also see that cooked_time (i.e., sys_ts+correction) is fed into > > SPF_AccumulateSample as sample->time. Later on, combine_selected_samples > > makes use of ->time to calculate various things. So, it looks like any > > error due to ns->us truncation will affect the math here. At a glance, I > > can't tell if the effect is so small that it can be ignored or if there is > > any benefit to avoiding the [0,999] ns error. > > It's so small that it can be ignored. The time at which the offset was > measured is not that important. The frequency error between the > reference and current clock is minimized by the control loop, usually > to less than a ppm. Even if it was 1000 ppm, over a 1 microsecond > interval that would still make only 1 nanosecond of phase error. It's > well hidden in the noise. The main purpose of that timestamp is > detection of stale and out-of-order samples.
Makes sense. Thanks for the context. Jeff. -- To unsubscribe email chrony-dev-requ...@chrony.tuxfamily.org with "unsubscribe" in the subject. For help email chrony-dev-requ...@chrony.tuxfamily.org with "help" in the subject. Trouble? Email listmas...@chrony.tuxfamily.org.
