[LinuxPPS] Ultimate time server

[Paul Lavender]

Thanks Javier for your info on Rubidium clocks, and others for your advice.

One thing I have picked up on is that a GPIO might be a better option 
than an RS232 DCD for the PPS

I haven't totally given up on the idea of feeding the motherboard clock 
externally, after all there were suggestions for changing the type of 
crystal, which involves the same amount of risky soldering!

  On 29/12/10 11:12, Javier Herrero wrote:
> Yes, I'm aware of the work of Poul-Henning :) He is also the author of
> the ntp Oncore clock code. And that is the way I've been playing with,
> hardware timestamping using an external counter (with much more
> resolution... 2.5ns). But when converting the timestamps from counter
> values (that have a jitter of only a few pps, due to the GPS pps output
> jitter - this jitter is later reduced by the ntp Oncore clock driver
> using the correction data provided by the GPS) to time values, I've
> found that a jitter appears, in the range of 300-400ns, and sometimes
> more. I suppose that due to other granularities in the kernel that I've
> not yet investigated.
>
> And the results that I obtain are good... but could be better :) My next
> steps will be to use a GPS-disciplined clock source for the timestamping
> and timekeeping, and to look were that jitter is appearing. But january
> will be a very busy month for me and I don't know when I will have time
> to continue playing with these things :)
>
> Regards,
>
> Javier
>
> El 29/12/2010 11:52, Gerhard Bertelsmann escribió:
>    
>> I am reading your interesting discussion about a precise time server. Poul
>> Henning-Kamp
>> from the FreeBSD universe has done some interesting work using hardware
>> time stamping
>> - see link below:
>>
>> http://phk.freebsd.dk/soekris/pps/
>>
>> The AMD Elan SC520 CPU has a kind of one-shot counter with a resolution of
>> 33Mhz/4
>> which could time stamp external events with a resolution of 120ns to get
>> rid off interrupt
>> latency/jitter. Poul has also used counters build on FPGAs on standard PCs
>> an achieved
>> amazing results. IMHO that is the way some of you try to archive high
>> accurate time servers.
>>
>> The Elan CPU is a little bit outdated nowadays. But there are new
>> interesting ones:
>> PowerPC derivates
>> Xscale (ARM)
>> and Marvell Kirkwood (88F6281 ARM)
>> Especially the last one is IMHO the most interesting one. It can do IEEE
>> 1588 and PTP external
>> events time stamping (e.g. GPS PPS) with a resolution of 8 ns. Andy
>> o'Brien has done
>> some interesting investigation. He was so kind to share his code. Please
>> have a look at:
>>
>> http://lnxpps.de/openrd_utils.tar
>>
>> The hardware used is OpenRD. Im trying to do this on a cheap Seagate
>> Dockstar which
>> is based on the same CPU. There are a lot of other boards, e.g.
>> SheevaPlug, QNAP TS-119, ...
>>
>> The code is not ready to use, but it should be a very good starting point
>> especially gpio9_irq_test.c.
>>
>> Somebody interested in making a PPS driver ?
>>
>> Regards
>>
>> Gerd
>>
>>
>>      
>
> _______________________________________________
> 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
>