[time-nuts] Re: NTP Accuracy

[Martin Burnicki]

Hi,

martyn at ptsyst.com wrote:
> Hello,
> 
> I have a board that is a NTP server.  Any ideas on how to test its accuracy?
> 
> To check it roughly I sync my PC to it, maybe set the time to +3 hours and
> confirm my PC follows.
> 
> But its supposed to be accurate to 500 ps and I want to verify that.

The easiest way to *measure* the accuracy would be to measure a hardware 
output signal (e.g. a 1 PPS slope, if the device provides one) against 
the 1 PPS signal of another device with known accuracy.

However, if you keep in mind that most commonly used GPS/GNSS receivers 
have an accuracy no better than a couple (or tens of) nanoseconds, so it 
will probably be hard to verify 500 ps accuracy.

Generally, even if the internal time base really has an accuracy of 500 
ps, you won't be able to get this level of accuracy into your client 
using the NTP protocol because NTP has to cope with a bunch of unknown 
latencies between the server and the client. See, for example:
https://kb.meinbergglobal.com/kb/time_sync/time_synchronization_accuracy_with_ntp

The PTP protocol supports hardware timestamping of PTP network packets, 
so if *each* node of the underlying network hardware also supports this, 
you can yield much better accuracy at the client side.

In any case, the accuracy you can yield at the client side depends 
strongly on the time synchronization software, the operating system, and 
the hardware of the client. See the chapter "Time Server Tasks vs. Time 
Client Tasks" of the KB article mentioned above.

Martin
-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature.sig
Type: application/pgp-signature
Size: 840 bytes
Desc: not available
URL: <http://febo.com/pipermail/time-nuts_lists.febo.com/attachments/20220221/d8db2e91/attachment.sig>