On Wed, 9 Jun 2004, Matthias Drochner wrote:

> fredb@immanent.net said:
> > Please try the jumbo patch I posted to PR 23929.
>
> I wasn't able to extract a working patch from the PR,
> so I've just given the latest snapshot
> (ntp-stable-4.2.0a-20040608) a try.
> ntpd seems to behave more peacefully than 4.2; it settles
> quickly, without the overshoot observed before.
> Sadly, for IPv6 they did neither follow the NetBSD patches
> (which I had sent to their bugzilla) nor the current (bind9)
> libisc and came out with stuff which doesn't work. So we'll
> have to maintain our local fixes for a while.

I'll try to keep up, so that the next snapshot may hopefully
build and run on NetBSD.

> I'll be traveling for the rest of the week; I'll do some
> more testing next week and eventually update the in-tree
> version.
> Could you put your jumbo patch somewhere or send it as
> attachment -- it might be preferrable to apply it first
> to get something easily pulled into the 2.0 branch.

That was my idea, too. Here are some modest changes, as of
February. This is probably the best candidate for committing and
pulling up to 2.0. The chief innovation seems to be that the poll
interval goes very low after a step, minimizing the time spent at
the mercy of the entropy due to few samples, and therefore the
absolute damage. That, and a few pre-requisites which permit the
Bitkeeper patch to pull up cleanly.

 ftp://ftp.netbsd.org/pub/NetBSD/misc/fredb/ntp-poll_interval.diff

Following is the same, plus Dave Mill's complete rewrite of the
loop filter committed to ntp-dev on March 2. It's already out of
date, though, so I don't think it would make sense to commit this.

 ftp://ftp.netbsd.org/pub/NetBSD/misc/fredb/ntp-superpatch.diff


Frederick