[LinuxPPS] gpsd effect, was Re: latency spikes, was...

[gnu not unix]

In message <45224102.2060809 at xs4all.nl> you write:

>Udo van den Heuvel wrote:
>Did somethign change in the kernel. I am runnign 2.6.18 since a few
(...)

Yes, the post 2.6.18 kernels have made Very Extensive changes in the
areas that ntpd is concerned with. I had a drifty time with the
short run I made with 2.6.18 and went back to the 2.6.17 patch here.

I've had a little experience with the gpsd code running with ntpd now.
It turns out that gpsd enables several extra NMEA strings beyond just
the $GPGGA that I use with clock type 20. 

These extra strings caused the linuxpps offsets to flap up to 40 (!)
microseconds compared to the 5 microseconds with only the one NMEA
string.

Here's some plots of a run with gpsd compatible output strings
on a gps35 and linuxpps kernel. 

http://www.wraith.sf.ca.us/ntp/linuxpps-gpsd.html

The freebsd and linux PPSKit kernels see some additional noise
in their ntpq data plots, but do not seem to suffer the degradation
in offsets that linuxpps saw during this interval. I am wondering
if some additional filtering in the linuxpps driver will provide
similar performance to the other two kernels.

../Steven