[LinuxPPS] LinuxPPS for kernel 2.6.26

[Hal V. Engel]

On Monday 14 July 2008 09:00:15 am Remco den Besten wrote:
> OK, indirectly thanks to Regs latest post I think I got it running with my
> contrapsion.
>
> Apperantly I have to use a tailing ampersand to load either ppsldisc or
> ttyctrl.
>
> When I enter (sudo) ./ttyctrl -X -s 18 /dev/ttyS0 & things seem to work.
> Directory pps0 is made and I can see activity on pps0 with:
> watch -n1 cat /sys/class/pps/pps0/{clear,assert}
>
> I have to wait a while to see the quality (improvement) of this latest
> LinuxPPS implementation.

With the 2.6.26-rc8 kernel and ntp built with the correct headers for 
nanoseocnds I have noticed that ntp does act somewhat differently.  Mostly it 
makes smaller (finer) adjustments to the clock frequency than it did when 
running with a microkernel.  The result is that it takes it longer to deal 
with things like temperature changes that affect the clock ocillator.  When 
the clock is affected by this sort of thing it will end up with larger offsets 
before ntp starts to pull it toward 0 again then would have been the case with 
a microkernel under the same conditions.    It also takes it longer to 
stabilize the clock after a cold start.

I suspect that a nanokernel setup would work better with a very stable clock 
(like a PC that has been modified by replacing the XO with a TCXO) but may 
actually not work as well as a microkernel if your clock is not very stable 
(like most unmodified PCs which are about +-1ppm/degree C).

In addition, on my machine ntpq -p or ntpq -c rv reports higher levels of 
jitter than it did with a microkernel.  WIth the microkernel it would 
typically report jitter as 0.001 once the clock was stable but with the 
nanokernel it is 0.002 under the same conditions.  I suspect that this is a 
truncation (microkernel - since values were limted to microsecond resolution) 
vs. rounding (nanokernel) type of thing. 

When I run ntpq -c rv I notice that the reported precision is either -19 or 
-20.  Which is 2 us or 1 us respectively (why do I get different values?).  
This has not changed since moving to a nanokernel.  I am not sure why since I 
have expcted that with the nanokernel it would report something like -30 which 
would be nanosecond precision.  Does anyone know what is going on with this.  
Is this an issue with ntpq? ntpdc kerninfo reports precision as 1e-09 s which 
is nanoseconds but ntpdc sysinfo reports precision as either -19 or -20.   
ntptime reports that precision is 0.001 us which is nanoseconds.  Is there 
someplace where this is documented?

>
> And, Rodolfo, I forgot:
> Yes, I did compile a new ntpd against the latest rc8 patch (forgot to
> mention it earlier ;-)
>
> I'll keep you informed here.
>
> Remco
>
>
>
>
> _______________________________________________
> LinuxPPS mailing list
> LinuxPPS at ml.enneenne.com
> http://ml.enneenne.com/cgi-bin/mailman/listinfo/linuxpps
> Wiki: http://wiki.enneenne.com/index.php/LinuxPPS_support

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://ml.enneenne.com/pipermail/linuxpps/attachments/20080714/9ad77d00/attachment-0001.htm