| Re: [chrony-users] chrony not observing maxpoll - bug? |
[ Thread Index |
Date Index
| More chrony.tuxfamily.org/chrony-users Archives
]
William G. Unruh __| Canadian Institute for|____ Tel: +1(604)822-3273
Physics&Astronomy _|___ Advanced Research _|____ Fax: +1(604)822-5324
UBC, Vancouver,BC _|_ Program in Cosmology |____ unruh@xxxxxxxxxxxxxx
Canada V6T 1Z1 ____|____ and Gravity ______|_ www.theory.physics.ubc.ca/
On Fri, 5 Feb 2021, Charles Laub wrote:
Background: I need to keep several machines well synchronized to my local NTP reference. This is for audio work, where playback needs to be tightly synchronized
between multiple computers. I previously used NTP for many years for this, but have recently switched to chrony to take advantage of its time smoothing and
filtering capabilities. I have a local "server" in my basement where the temp remains constant. It's a stable reference for other computers in my home. My
machines only poll only this one sever.
On various Linux clients, I am experiencing the following issue:
Chrony works as expected. Synchronization is achieved to high precision, e.g. tens to few hundred microseconds.
I use a fixed polling interval by setting minpoll and maxpoll to e.g. 2
Not a great idea. This will mean that the chrony has a hard time figuring out
what frequency it is supposed to run at. (it will be noisy.) There is an
optimal poll interval at which both the offset and the frequency are "best",
and it is a lot longer than poll 2.
To keep track of what is going on I print out the last offset every 60 seconds in an ssh terminal session
I have noticed sometimes that this value gets "stuck". Here is some example output:
That is not stuck. That is an oscillation. If you let the poll interval be
longer I suspect you would get much better behavious. Which a bunch of
computers in your own house, with a reasonable connectivity (and I would
suggest wired connection between the machines, not wireless) you should get
more like a few microseconds after it settles down.
Last offset : +0.001239868 seconds
Last offset : -0.001022636 seconds
Last offset : +0.000342921 seconds
Last offset : -0.000313532 seconds
Last offset : -0.001234458 seconds
Last offset : -0.001234458 seconds
Last offset : -0.000906603 seconds
Last offset : -0.000906603 seconds
Last offset : -0.000036374 seconds
Last offset : -0.000354126 seconds
Last offset : -0.000354126 seconds
Last offset : -0.000010771 seconds
Last offset : -0.000010771 seconds
Last offset : -0.000010771 seconds
Last offset : -0.000010771 seconds
Last offset : -0.000411757 seconds
Last offset : -0.000411757 seconds
Last offset : -0.000411757 seconds
It dawned on me later that this should not be happening, and when I looked at the chronyc output I found this:
pi@SPKR-left:~ $ chronyc tracking
Reference ID : C0A801FE (192.168.1.254)
Stratum : 4
Ref time (UTC) : Fri Feb 05 18:50:42 2021
System time : 0.000003940 seconds fast of NTP time
Last offset : +0.000011845 seconds
RMS offset : 0.000603218 seconds
Frequency : 6.857 ppm fast
Residual freq : +0.000 ppm
Skew : 0.059 ppm
Root delay : 0.039830066 seconds
That is huge in your own house.
Root dispersion : 0.003487853 seconds
Update interval : 535.4 seconds
Leap status : Normal
Notice the reported update interval of 535 seconds. But my server line includes the directives:
minpoll 2 maxpoll 2 iburst filter 8
The poll interval should remain at 2, and with filter=8 the poll should be reported every 32 seconds. 535 is certainly not the same as 32, so I suspect a bug.
This machine (Pi 4) is running chrony version 3.4, and another (an Intel box) is running 3.5. Both show an increase in the poll interval above 32 seconds in
chronyc. When this happens, the synchroncity becomes relatively poor so I would like to find a way to fix this problem if possible.
Let me know if there is some other info I can post that might be useful.
Thanks,
-Charlie