On Sun, 6 Jun 2004, Tom Ivar Helbekkmo wrote:

> Since a couple of months ago, I've seen problems with ntpd.  All my
> Intel boxes running NetBSD-current now look like their clocks are too
> poor for ntpd to maintain them -- but until a couple of months ago,
> they all worked just fine.  One machine looks like this:

No doubt, the change was conincident with the import to current of NTP
4.2.0, from 4.0.99i.

> Both of these are configured with several known good ntp servers.
> Note that the resets go both ways, which doesn't rhyme with a poorly
> adjusted clock.
>
> Any hints as to how to configure these boxes to have more stable clocks?

Please try the jumbo patch I posted to PR 23929. I understand that
that code, from ntp-dev, will be in NTP 4.2.1. The loopfilter seems to
behave much better with that in the face of brief periods of
congestion or missing samples, for me, but there's still a problem
with SMP that obscures the benefit somewhat, and of course it takes a
lot of time to really judge, so I'm looking for some positive feedback
before committing. Thanks.

Frederick