Subject: Re: word that isn't !h!e!l!p!: xntpd (2nd)
To: henning loeser <loeser@ma1304.physik.uni-marburg.de>
From: Henry B. Hotz <hotz@jpl.nasa.gov>
List: port-mac68k
Date: 02/06/1998 11:46:21
At 4:10 AM -0800 2/6/98, henning loeser wrote:
>I fixed that and now Xntpd sycs to the server, loses sync , syncs again,
>but after a while it stopps doing that. I guess it's when I start
>xlockmore. (I start it with nice +19, hoping that this way it wouldn't
>cause too much of a load if anything else needs computation time.) Does

Unfortunately the nice value won't have much effect because the priorities
in question are all inside the kernel interrupt handlers.  Sorry |-(

Some things you can try:

Delete the ntp drift file to start from scratch, like I said last time.

Run the machine without ntp for a few days to calibrate the clock drift
rate against a good source.  Then use that drift rate to initialize the
drift file by hand.  (Don't remember the units.  See the ntp docs, or
http://www.eecis.udel.edu/~ntp/, for details.)

Try playing around with the fudge command to change some of the loop
parameters.  The standard parameters make some assumptions about the
reliability of local clocks compared to remote ones that MacBSD can't meet
due to the probability of loosing clock interrupts.  In theory you should
be able to tune the loop parameters so ntp is more likely to trust the
remote clock and to just fix the local one.  Again look elsewhere for
details, and if you haven't a clue about phase locked loops then you have a
bit to learn.

We had a discussion here a long time ago about perhaps running ntp off of
the once/second interrupt.  In essence you need a device driver for that
interrupt that you can feed to ntp as a 1pps reference (so you need an ntp
driver to use it as well).  The idea is that the 1pps interrupt is a lot
less likely to be lost than the 60Hz interrupt used for the kernel clock.
This idea may not work without also doing the above tuning as well, but it
would be a *big* help for people without a continuous Internet connection.

Finally you could just do what I'm told puma does and run ntpdate from cron.

Signature failed Preliminary Design Review.
Feasibility of a new signature is currently being evaluated.
h.b.hotz@jpl.nasa.gov, or hbhotz@oxy.edu