[LinuxPPS] refclock_nmea patch

[Bernhard Schiffner]

Am Sonntag, 28. Oktober 2007 09:09 schrieb Johan Bejeryd:
> Hi,
>
> I've tested the patched NMEA driver with the kernel from the git
> repository and it's working; however the convergence  is  slow.  It took
> about 8 hours for the offset to drop from 1500 us to 5 us.  Are you
> experiencing the same results?
>
> /Johan

Yes,
that's the experience. But: it's no must have. With 2.6.19 the 
kernel/time/ntp.c changed into it's "modern" shape. I got this (very) slow 
ntp behavior after reboot first (ntpd was running before and in good 
shape) see

http://www.schiffner-limbach.de/ntp/ntp_logs_061210.png
http://lkml.org/lkml/2007/1/10/75

Somesimes it happens, sometime not.
In general now ntp is slower then it was before. I have had a look at the 
sources but couldn't chatch the reason / difference. 

Johan, if you could collect some data more about this please?
If this happens in a repeatable way, it makes IMHO sense to start a 
discussion with the ntp-people (linux and others) about the time a ntpd 
(with PPS!) is allowed to catch-up.
In an embedded environment it makes no sense to have PPS but adjustment 
takes half a day.

Bernhard

(The diagram was generated by gnuplot with a little script around. Want to 
see it?)