> From: "Charles M. Hannum" <mycroft@gnu.ai.mit.edu>
> Subject: Re: Why I wanted to tweak tickadj
> 
> 
>    It may not be, in general.  In my case, I had an adjustment large
>    enough that I didn't want to wait for the slow one-or-two-percent slew
>    tickadj tends to use; I cranked tickadj up until the machine's
>    time-of-day was ticking off seconds nearly twice as fast as my watch.
> 
> There's a good reason for `ntpdate' being in the xntp distribution.
> In general, you should run it before you start xntpd.
> 
>    One thing on my
>    it-would-be-nice-to-do-someday list is to look at extending things so
>    that the clock tick size is kept in units much smaller than
>    microseconds (perhaps 2^-32 usec) so that it can be adjusted to very
>    fine precision, to match observed effective clock tick rates quite
>    accurately.
> 
> There's a piece of code in recent xntpd distributions that's supposed
> to simulate a phase locked loop.  Supposedly after the PLL settles it
> keeps very precise time.  I haven't tried it.

There is a kernel PLL driver that is available for NTP. I'm planning on
trying to integrate it into NetBSD over the holidays. It's available
as kernel.tar.Z on louie.udel.edu (/pub/ntp?)