[time-nuts] Re: Poor's man NTP

[giuseppe at marullo.it]

Hi Hal,
Thanks for your answer.

>This is the time nuts list.  You have to decide how much time you want to
spend tuning, tweaking, and understanding NTP quirks.  It's a bottomless
pit.
I have a couple of GPSDOs, but my time nuttery is pretty limited, just
wondering if this is doable.

I see that NTP is a whole universe in itself, adding the "unusual" raspberry
serial port (NMEA sentences on one serial port BUT CD not available for PPS
signal) to the mix would take even more time to master.

My reasoning is that if Microsoft says 1ms accuracy is possible:

https://docs.microsoft.com/en-us/windows-server/networking/windows-time-serv
ice/configuring-systems-for-high-accuracy

I could/should have less than 1ms at the Stratum-1. Ntpstat command says 2ms
then something is wrong:
1) I am off by about 2ms
2) ntpstat does not report accuracy reliably
3) both


>> 0.europe.ntp.or .POOL.        ...
>I think you have a typo in there.  0.europe.ntp.org doesn't exist.  Try
0.europe.pool.ntp.org
Thanks and corrected, I also removed the server references, just pools.
Rebooted and waiting if this would improve anything. So far, no improvement.

>> version="ntpd 4.2.8p12 at 1.3728-o (1)",
>That's the classic Mills code.  Their support lists are broken.
>I suggest switching to NTPsec.  Debian has a ntpsec package.  Their lists
are working.  (Of course, as I type this, the server is down.  :)
> https://lists.ntpsec.org/
>
>Disclaimer: I work on the NTPsec software.

I appreciate the info but I would not start again from scratch with NTPsec,
this is the first setup that somewhat works and do use PPS signal. 
There are a lot of tutorials for doing the NTP Server+Rspberry recipe, but
either they are obsolete or inaccurate at best.
The one I found allowed me to skip GPSD at all, I was not able to make gpsd
work with my board.
Nevertheless I could give it a shot, like using this tutorial:
https://www.ntpsec.org/white-papers/stratum-1-microserver-howto/#_clockmaker
Do you have any better one to use?

-------

>> BTW I don't know why reftime and clock time differs so much, and I 
>> don't know why the difference vary, from some seconds to several 
>> seconds if check even within a short time span.

>I'd have to check the code.  I think reftime is the time it last tweaked
the system clock.
You  do seem right. Difference is never more than 8 seconds, the time I
configured as polling interval.




>I can't read Italian.
> I usually set things up with separate slots for the NMEA device and the
PPS device, just so I can get more information.
Don't worry, main thing is that is not using gpsd, everything else is pretty
standard. I got the serial port reversed (prolly because I switched the SD
from a pi-zero to the PI4), other than this it worked.