Re: NTP via tcp NMEA feed

[James Browning]

> On 10/14/2022 5:16 AM PDT Nick Taylor <nicktaylor@dataskill.uk> wrote:
> 
>  
> Hi guys
> 
> You may have seen that we are working with using tcp NMEA feeds for a 
> device that we built - basically certain customers have existing GPS 
> devices and wish to use feed from that instead of separate antenna...
> 
> Additionally often Internet connectivity is firewalled back so we drop 
> back to using GPS time rather than proper NTP
> 
> Now I notice that I don't seem to see NTP coming through from NTP feed - 
> the only success we had was using the shm feed and with normal setup 
> ntpshmmon shows normal time feed coming through nicely which we can then 
> link into chrony.
> 
> Is this expected behaviour or a bug? Is there any way to still get NTP 
> from a tcp NMEA feed??
:::snip:::

You would need something that speaks the NTP protocol. No, scratch that; you 
should probably use something that talks PTP. NTP typically has this annoying 
requirement that you have multiple servers for everything unless you hack 'tos 
minsane' and 'tos minclock' values. I expect the Chrony has similar adjustable 
values.

I could write something reasonably quickly that serves *the wrong* NTP time. To 
do it correctly, you would need UTC, GPS, and local machine time at (roughly) 
the exact moment. As well as some logic to fill in some values that SNTP 
doesn't use. The IERS bulletin C file and tzdata database would probably be 
helpful, as would the clock on a smartphone.

I could joke about smartphones having most (all) of the internal bits needed 
for a good stratum one time server except for a pulse-per-second from the GPS 
receiver.