[LinuxPPS] Linux 2.6.28 - serial time string jumps

[Heiko Gerstung]

Hal V. Engel schrieb:
> On Thursday 15 January 2009 06:29:47 Heiko Gerstung wrote:
>   
>> OK, new day, new problem...
>>
>> Warning: This might be a little bit offtopic i.r.t. linuxpps, but I have
>> seen that you guys are dealing with bleeding edge versions of the
>> kernel, linuxpps and ntp (just like me) and might have similar
>> experiences...
>>
>> I am trying to test the new kernel with Rodolfo's linuxpps patch against
>> ntp-4.2.4p6 ... PPS now works reliably but what I see is that the time
>> of my GPS string reference (parse driver) is sometimes jumping for 3-4
>> ms, resulting in being called a bad ass falseticker by ntpd and striked
>> out (x). This in turn leads to the PPS source being rejected and after a
>> short while (I am dealing with 16s polling intervals) everything seems
>> to kick back to normal ...
>>
>> The same hardware performs good under 2.6.15.1 (the last kernel version
>> I was working on), back then with the PPSLight patch from Ulrich Windl.
>> While I assume that PPS is working great here, I wonder if anyone has
>> experienced jumps and latency issues on their serial ports with 2.6.28?
>>
>> I know that I had some successful tests with 2.6.24 but I wanted to use
>> a more recent kernel ...
>>
>> Regards,
>>  Heiko
>>     
>
> Have you tried: 
>
> setserial /dev/<your port> low_latency
>
> This should make the latency of the serial port lower and also more 
> consistent.   I have not seen this issue with any of the machines I have this 
> on.
>   
This seems to help a lot, thanks. I still face problems regarding the 
speed of convergence and currently I am testing with minpoll 1-3 
(required a little patch for ntpd), which seems to work fine. One thing 
I am still seeing is that I have occasional negative spikes of up to 50 
microseconds. This happens around 3x a day and in my offset graph I see 
a +/-3us offset and then a sudden jump to -50 which is immediately 
disappearing and everything goes on as normal.

BTW: Has anyone a link for me that describes which kernel config 
settings affect timing? There are numerous things and I guess that 
almost everything counts when we are at the lower microsecond level, but 
I wonder if anyone went through this before and noted which settings are 
affecting the overall (ntpd) timing performance.

Regards,
Heiko

> Hal
>
> _______________________________________________
> 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
>