4.2.1-RC1 on FreeBSD 4.11 doesn't seem to discipline the local clock

4.2.1-RC1 on FreeBSD 4.11 doesn't seem to discipline the local clock

Post by Joshua Coo » Thu, 09 Mar 2006 13:43:41


I've got a Soekris 4501 (three actually) running FreeBSD 4.11. I just
added a modem to pickup ACTS with, and discovered that ntp in base
FreeBSD isn't compiled with it as a valid refclock? So, I pulled down
the 4.2.1 release candidate, configured with just --enable-acts, and
gave it a shot. The clock just slowly drifted away from it's peers
and ACTS. Up to this point, using base ntpd from FreeBSD and multiple
internet sources (and it's 3 local peers I maintain) it has stayed
within 10ms, I cut the cord when it drifted out past 150ms this
evening.

Should I be using additional configure arguments on FreeBSD 4.11?

Joshua Coombs
 
 
 

4.2.1-RC1 on FreeBSD 4.11 doesn't seem to discipline the local clock

Post by Harlan Ste » Thu, 09 Mar 2006 16:15:33

What does 'ntpq -p' show on that host?

H

 
 
 

4.2.1-RC1 on FreeBSD 4.11 doesn't seem to discipline the local clock

Post by Joshua Coo » Thu, 09 Mar 2006 23:22:03


I've just reset the time via ntpdate -b, and resumed using 4.2.1-RC1

remote refid st t when poll reach delay offset
jitter
==============================================================================
ntsrv04.jensenr .CDMA. 1 u 39 64 3 17.045 0.005
0.479
tick.jrc.us toshi.keneli.or 2 u 37 64 3 18.125 -2.217
0.418
tock.jrc.us NAVOBS1.MIT.EDU 2 u 35 64 3 16.927 -0.486
0.388
node-217.machin .USNO. 1 u 34 64 3 25.576 0.017
31.221
bonehed.lcs.mit .CDMA. 1 u 32 64 3 17.887 -0.891
0.290
toshi.keneli.or .GPS. 1 u 33 64 3 23.906 -0.225
0.326
clock.isc.org .GPSi. 1 u 30 64 3 83.531 0.071
0.665
clepsydra.dec.c .GPS. 1 u 31 64 3 82.202 -0.983
0.359
c-66-229-200-20 .GPS. 1 u 31 64 3 88.131 -2.912
0.550
timekeeper.isi. .GPS. 1 u 28 64 3 110.157 9.323
14.243
clock.xmission. .GPS. 1 u 27 64 3 72.805 -4.671
0.648
rassilon.gwi bonehed.lcs.mit 2 u 62 256 2 5.926 0.391
0.015
omega.gwi toshi.keneli.or 2 u 15 1024 3 6.768 -0.559
0.155
salyavin.gwi .INIT. 16 u - 64 0 0.000 0.000
0.000
azmael.gwi ntsrv04.jensenr 2 u 40 1024 2 1.298 -0.193
0.015
LOCAL(1) .LOCL. 10 l 24 64 3 0.000 0.000
0.015
*ACTS_NIST(1) .NIST. 0 l 23 2048 1 0.000 -5.629
3.916

Rassilon and Azmael are both identical 4501s, matching installs of
FreeBSD, running the base included NTPd (4.1.0-a). Omega is the 4th
peered server in this group, a Cisco 4500M running 12.3 IOS. I use
them as my comparison clocks to see if changes on Salyavin (the 4501
I'm testing 4.2.1-RC1 on) cause problems.

I also forgot to create a dir for my loopstats, so I don't have prior
ones to reference. I've corrected that and am gathering stats now if
those will help.

One other change in behavior I've noticed, after about 3 rounds of
polling at 64 seconds, ntpd jumps right up to 1024 seconds for a
polling interval. I personally like the idea that I'm less abusive to
remote servers, but just want to make sure it's proper behavior and
not something that should be reported as an issue?

Joshua Coombs
 
 
 

4.2.1-RC1 on FreeBSD 4.11 doesn't seem to discipline the local clock

Post by Harlan Ste » Fri, 10 Mar 2006 16:35:36

The output you sent us seems to have been from a "fresh" start of ntpd, as
the "reach" was never more than 3.

Why do you think your locak clock was not being disciplined?

H
 
 
 

4.2.1-RC1 on FreeBSD 4.11 doesn't seem to discipline the local clock

Post by Joshua Coo » Sat, 11 Mar 2006 00:33:58


Correct, I had reverted to 4.1.0a when I first noticed the huge
offsets. Letting 4.2.1-RC1 run yesterday it appears that it is in
fact discipline the clock, but can't seem to settle to stable
operation. The clock would begin to run off in one direction, ntpd
would lower the poll intervals, over do the correction value, then
raise the poll intervals while the clock came back to 0 offset. The
pattern would repeat as it swung past in the other direction. Running
4.1.0a, ntpd very quickly reigns in the offset, and then stays within
10ms of true time without the drama and poll interval ramps of
4.2.1-RC1.

The huge swing initially is what made me fear it wasn't discipling the
clock, letting it run longer shows that it is, but for some reason
can't find a happy stable state.

I'm setup to collect loopstats now, and because I have 3 other clocks
that are stable I can use this machine as a guinea pig for a bit, how
should I proceed to diagnose this?

Joshua Coombs
 
 
 

4.2.1-RC1 on FreeBSD 4.11 doesn't seem to discipline the local clock

Post by Harlan Ste » Sat, 11 Mar 2006 04:15:43

>>> In article < XXXX@XXXXX.COM >, "Joshua Coombs" < XXXX@XXXXX.COM > writes:

Joshua> Letting 4.2.1-RC1 run yesterday it appears that it is in
Joshua> fact discipline the clock, but can't seem to settle to stable
Joshua> operation. The clock would begin to run off in one direction, ntpd
Joshua> would lower the poll intervals, over do the correction value, then
Joshua> raise the poll intervals while the clock came back to 0 offset. The
Joshua> pattern would repeat as it swung past in the other direction.
Joshua> Running 4.1.0a, ntpd very quickly reigns in the offset, and then
Joshua> stays within 10ms of true time without the drama and poll interval
Joshua> ramps of 4.2.1-RC1.

Joshua> The huge swing initially is what made me fear it wasn't discipling
Joshua> the clock, letting it run longer shows that it is, but for some
Joshua> reason can't find a happy stable state.

Joshua> I'm setup to collect loopstats now, and because I have 3 other
Joshua> clocks that are stable I can use this machine as a guinea pig for a
Joshua> bit, how should I proceed to diagnose this?

This is an area where I have done very little, so I don't know.

I do know that http://www.yqcomputer.com/
be a very good place to discuss this.

H