Joshua Coombs wrote:
My nanobsd testbed has decided to drive me up a wall. When I set it
loose, it starts hunting for a pll adjustment value, swinging wider and
wider with each period. When I last restarted ntpd to try a conf
change, it was varying the pll freq by 40ppm over the course of one of
it's sweeps.
Plotting the adjustment value over time generates a lovely sine wave
with a static period but an exponentially growing amplitude centered on
-25ppm.
The box is running FreeBSD 6.1B4, and I've alternated between the base
included ntpd (4.2.0a) and the 4.2.1RC and both show the same behavior.
I was initially locking the poll interval at 2048 seconds as this is a
test setup and I both wanted to verify the stability of the 4501 and
keep from abusing the remote servers I was using. I've lowered it to
512 seconds to see if that improves things...
Any suggestions on where to look for the cause? I picked 4501s based on
PHK's write-ups covering their excellent timekeeping abilities, and am
not seeing it in practice so I'd like to know where I err'd.
Joshua Coombs
Probably by locking the poll interval. There are excellent reasons,
involving stability, behind the default values for MINPOLL and MAXPOLL.
Configuring NTP with the default values is not considered server
abuse. Server abuse is polling at 1 second intervals!!
If you try the defaults, your box should synch up nicely and increase
its poll interval to the value that gives the best stability. If you
are using a good server and a good network connection it should probably
go to 512 or 1024 seconds and stay there as long as conditions don't
change.
You neglected to say what you are trying to synchronize with and that
may be relevant. FWIW, PHK was using GPS reference clocks on his
Soekris boxes.
_______________________________________________
questions mailing list
[email protected]
https://lists.ntp.isc.org/mailman/listinfo/questions
