2.6.19-rc2 and very unstable NTP

From: linux
Date: Thu Oct 26 2006 - 05:24:50 EST


There have been a bunch of changes to timekeeping since 2.6.17, and I've
noticed that, with 2.6.19-rc2 + linuxpps, I'm getting some impressively
unstable oscillations in the local time. +/- 350 us, when, given a
good quality local PPS source, it should be wiggling +/- a few us.
And the shape of the curves is not "wander" but "overcorrecting wildly".

It's sort of series of exponential decay curves, but each one overshoots
by 100%, and then before it fully flattens out, starts surging in the
other direction. (Actually, the overshoot amplitudes fluctuate
erratically, too.) As best I can render it in ASCII art:


* * *
* *
* *
* *
* * * *

* *
* *
--------------*----------------------*-----------
* *

* * * *
* *
* *
* *
* *

600-700 us p-p, with 2200-2400 s per full cycle.

It looks like ajtimex() isn't doing what NTP is expecting,
leading to loop instability.


I'm going to git bisect this, although it's a bit time-consuming
waiting to see if things will settle own cleanly after each reboot.
And some of the patches have been anything but one-liners, so
that's not necessarily a direct pointer to the problem.

I know there have been a number of reports of ntp timekeeping
problems with 2.6.18. Has there been any progress already?

(Local system: AMD64 uniprocessor, 2.6.19-rc2+linuxpps kernel,
NTP 4.2.2, Acutime 2000 GPS clock + PPS input.)
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/