[ntp:questions] stepping and slewing
martin.burnicki at meinberg.de
Fri May 11 07:25:19 UTC 2007
David L. Mills wrote:
> The NTP clock discipline doesn;t work like that. The ntpdate and
> imitator compute an offset correction, hand off the adjtime() and go
> home. The NTP clock discipine algorithm incorporates a feedback loop
> which includes a proportional control with time constant of several
> teens of minutes. See the arthictecture briefing at the NTP project page.
I have seen lots of astonished requests why it takes so long for ntpd to
adjust a small initial time offset of e.g. e couple of milliseconds.
Wouldn't it be possible and useful to start up with a smaller time constant
which allowed for faster initial corrections, and then increase the time
constant to the final value to move over to the normal behaviour of ntpd?
More information about the questions