Dave Hart wrote:
On Feb 7, 12:24 UTC, David Lord wrote:
Downside of not having fudge is first few NMEA offsets are 350ms
until the NMEA driver sorts out its own handling of pps. As soon as
offset shifts to a few ms or less jitter then rockets to near 350ms
and works down from there fairly quickly.
It's probably worth trying with the fudge time1 in place as that
might just improve startup.
I would suggest you upgrade to ntpd 4.2.6 or 4.2.7. Based on your
chronicles, I'm guessing you're using 4.2.5 or earlier, because you
note that the NMEA driver is switching from serial end-of-line
timestamps to PPS timestamps soon after starting, and because I'm
assuming you don't have a fudge flag set for that driver to enable its
PPSAPI use. IIRC, that is the classic behavior of NMEA compiled on a
system with PPSAPI support -- if PPSAPI works on the NMEA serial port,
it's used unconditionally. With 4.2.6, you have to specifically ask
the NMEA driver to enable PPSAPI use with fudge 127.127.20.x flag1 1.
Moreover, with older NMEA there is a single offset fudge available,
used at least for serial end of line fudging but I think it was also
used to offset the PPS timestamps. With 4.2.6, fudge time1 affects
only NMEA PPS timestamps, and fudge time2 affects only NMEA serial EOL
timestamps. That should enable you to cleanly fudge away the startup
350ms offset.
If you do go to 4.2.6 or later, I suggest you do not attempt to use
the same PPS signal via two different drivers at the same time, unless
one is marked noselect. That is, either use NMEA alone with fudge
flag1 1 to enable PPSAPI, or use NMEA without fudge flag1 and marked
prefer alongside PPS/ATOM (22).
I think you've explained/confirmed results I've had. I'm
using 4.2.4p6-o on the test box. I'm not likely to try install
of a more recent version unless it was going to solve problem
that I can't get parallel (or serial) pps to work with just
ATOM driver flag3 0 | 1 (kernel discipline disable | enable)
and no other refclock, ie I'd like to just use another ntp
server as preferred with pps to condition the clock.
cheers
David
_______________________________________________
questions mailing list
[email protected]
http://lists.ntp.org/listinfo/questions
