Re: [chrony-users] PHC and source selection

[Bill Unruh <unruh@xxxxxxxxxxxxxx>]

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 Thu, 22 Jul 2021, Ross Ilott wrote:

> [CAUTION: Non-UBC Email]Hello,
>
> I have a server running chrony, using a PHC refclock which is PTP-synced via ptp4l and also 3 NTP servers in its config. When my
> PTP source is not in-sync with GPS time it starts drifting, and you can see the effect of this in chrony on my server as follows
> (below is when it has drifted ~550us):
>
> 210 Number of sources = 4
> MS Name/IP address         Stratum Poll Reach LastRx Last sample
> ===============================================================================
> #* PHC0                          1   2   377     4    +17ns[   +5ns] +/-  523ns
> ^- asia1                            1   6   377    56   +559us[ +560us] +/-   43ms
> ^- asia2                            1   6   377    56   +565us[ +567us] +/-   50ms
> ^- asia3                            1   6   377    51   +565us[ +567us] +/-   50ms
>
> My question is, why does chrony not select a different source in this scenario? It has 3 other NTP servers that are perfectly
> good, and all seem to agree on the true time. Is it because the estimated error of the NTP servers are all ~50ms?

Probably.

> These NTP servers are geographically far away (100ms ping), but they themselves are synchronised from GPS sources, so their
> estimated error shouldn't be half of their network latency, they are much more accurate than that. Shouldn't the jitter of their
> measurements matter more than their latency?

No. For all the NTP algorithm knows that 50ms is 49ms upsteam and 1 ms
downstream. The only evidence it has is the total roundtrip time.

> As far as I can see there is no way to influence the estimated error of an NTP time source - is this correct?
> Ideally when my local PHC0 source is 500us away from the agreed time from the other 3 sources, it would get marked as bad.

But they are all 50ms away from the right time. Why would it want to pick
those?


> Thanks!
> Ross
>
> chrony tracking:
> Reference ID    : 50484330 (PHC0)
> Stratum         : 2
> Ref time (UTC)  : Thu Jul 22 16:22:59 2021
> System time     : 0.000000029 seconds fast of NTP time
> Last offset     : +0.000000001 seconds
> RMS offset      : 0.000000042 seconds
> Frequency       : 13.066 ppm fast
> Residual freq   : +0.000 ppm
> Skew            : 0.005 ppm
> Root delay      : 0.000000500 seconds
> Root dispersion : 0.000004934 seconds
> Update interval : 4.0 seconds
> Leap status     : Normal
>
> The config for chrony is like so:
> "
> makestep 1 3
> rtcsync
> hwtimestamp *
>
> driftfile /var/lib/chrony/drift
> logdir /var/log/chrony
> log rawmeasurements statistics tracking refclocks
>
> leapsectz right/UTC
> refclock PHC /dev/ptp0 poll 2 tai delay 500e-9 precision 1e-9 stratum 1
>
> server asia1 minpoll 4 maxpoll 6 iburst
> server asia2 minpoll 4 maxpoll 6 iburst
> server asia3 minpoll 4 maxpoll 6 iburst
> "
>
> and the ptp4l command:
> ptp4l -m -f /etc/ptp4l.conf -i nic0
>
> cat /etc/ptp4l.conf:
> [global]
> clock_servo linreg
> delay_mechanism E2E
> slaveOnly 1
> domainNumber 0