ublox refclock

Trevor N. trv-n at comcast.net
Mon Dec 2 23:15:33 UTC 2019

On Sun, 01 Dec 2019 15:16:39 -0500, Trevor N. <trv-n at comcast.net>

>>Trevor N. trv-n at comcast.net wrote
>>I have completed a 12-hour run, ntpviz plots are attached. It appears
>>that the on-board serial port (F9T) was showing some occasional jitter
>>spikes that were not as large as the ones on the pci serial port
>>(oncore). I'll do another run with a pci-e (on a CPU lane) serial
>>I forgot about ubxtool, so I'll reset the module to factory default
>>and then use it to modify the configuration with a script.
>Ntpviz data is attached for a 12-hour run using the pcie serial card.
>I have also included the ubxtool configuration and ntp.conf files.
> Mean SHM jitter decreased from 214ns to 146ns, but the local time and
>frequency jitter values are still worse than they could be.  I used a
>polling interval of 4, but I saw suggestions that performance can
>improve with polling set to 0.  I'll try again with a value of 0.

I completed a shorter run (6 hours) with polling set to 0. The mean
SHM jitter decreased to 142ns, but the max increased from under 900ns
to 3.1us.  It appears that the addtional data between polls is needed
for filtering on the machine I'm using.  At least with my current
setup, it seems that it's not possible to obtain better performance
with KPPS over the serial ports. I'm going to try again using the
parallel port PPS driver.  Could others using KPPS on fast machines
post some ntpviz charts?

After examining the oncore driver's source more carefully, I found
that I was mistaken about it not using sawtooth correction: The driver
has been using it since at least 2015, so it will be easy to remove
the correction and compare results.

I have added a post-timestamp-call 'echo' output on the DTR line to my
driver in order to take measurements with a time interval counter.
I'll update the issue 499 thread with the latest version.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: f9t-kpps-0poll-20191202a.tar.gz
Type: application/octet-stream
Size: 253598 bytes
Desc: not available
URL: <https://lists.ntpsec.org/pipermail/devel/attachments/20191202/ac5c0192/attachment-0001.obj>

More information about the devel mailing list