Hej Anders,
On 26/07/13 09:21, Anders Wallin wrote:
Hi all, I have two NTP-related questions:
1.
We are setting up a White-Rabbit[1] network for time-distribution. We
'seed' the WR-network with 10MHz and PPS signals form atomic clocks. This
means on each computer in the network there's a very accurate PTP-server
running on the WR-card, as well as the normal system time on the computer.
For fun I logged both the system-time (kept on time using NTP) as well as
the PTP time and plotted the error:
http://www.anderswallin.net/wp-content/uploads/2013/07/NTP_time_2013jul26.png
I was wondering if this plot is typical for a (good?) NTP-disciplined
computer clock?
There is folks here that have done tons of plots. You can do better I
think, but it depends on network, load and configuration.
Without NTP the free-running clock shows>40 ppm error:
http://www.anderswallin.net/wp-content/uploads/2013/07/freerunning_vs_ntp_2013jul26.png
The -16ms offset in this graph is probably due to my naive program where we
first call a simple function that asks the NTP time, and then over a serial
link ask for the PTP time. In reality the two time-stamps might be better
synchronized - I don't know. The slow variation I see should be real
however and completely due to drift in the NTP time, since the WR-time is
much more accurate.
2.
Is it possible to run several NTP-clients on one machine? That means I'd
have multiple "system-times" each synchronized to its own NTP server.
If this is possible I'd like to monitor several NTP-servers at once and log
their time-stamps against our WR-time which is known to be good.
Rather, you set up multiple peers to your NTPD and you then monitor them
as well as the system-time, and by combining source_n-NTPD and NTPD-WR
differences you get your full set of source_n-WR set of differences.
Looks like you have good reasons to tweak that NTPD setup, it should be
able to perform better than what you have logged.
Cheers,
Magnus
_______________________________________________
time-nuts mailing list -- [email protected]
To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.
