[time-nuts] verifying synchronization with PPS

[Majdi S. Abbas]

On Mon, Jul 08, 2019 at 09:59:52AM -0500, Graham / KE9H wrote:
> Several comments:
> 
> What level of accuracy do you mean by "synchronized"?
> 
> Plotting a lightly loaded Linux box, which is extracting time from the
> network via timesyncd, against a GPS 1PPS signal, I observe typical time
> excursions within +/- 10 ms, and extreme excursions over a 24 hour period
> of up to 20 ms.
> 
> So, I would not count on network based time on a Linux box to be better
> than +/- 20 ms, probably worse with heavy loading of the OS.

	Graham,

	timesyncd is at best an SNTP implementation that does not do
any real clock filtering, and latches onto its primary server only.
Unfortunately it's one of the things that came along for the ride with
systemd.

	Try disabling systemd-timesyncd, and configure either the
reference implementation ntpd, or chrony, and see where that gets you.

	On a good, quiet, symmetric link, I could generally keep a 
systems clock within hundreds of microseconds via NTP -- to peer 
clocks about 300 miles away.

	--msa