[time-nuts] Chooses for a desktop/server NTP external 1PPS reference

Bruce Griffiths bruce.griffiths at xtra.co.nz
Tue Dec 8 20:43:31 UTC 2009


Alexander Sack wrote:
> On Tue, Dec 8, 2009 at 3:01 AM, Robert Atkinson<robert8rpi at yahoo.co.uk>  wrote:
>    
>> Hi Alexander,
>> Welcome to the list.
>> Another professional option is the Trimble Accutime Gold. http://www.trimble.com/timing/  The older Accutime 2000 and Palisade turn up on ebay. A desk top box is the CNS Clock II http://www.cnssys.com/cnsclock/CNSClockII.php cost about $3000 complete.
>> Some idea on if you want new / professional or used / amateur and if you are a keen constructor would allow more targeted advice.
>>
>>
>>      
> Thanks Robert!
>
> Yeah I didn't really specify price or performance metrics in my
> request.  I think I should document this stuff and write up a
> beginner's FAQ (I didn't see anything on LeapSecond.com).
>
> Alright, the reason WHY I didn't actually specify much is because I
> really don't know WHAT to look for.  I found out today that the
> rinky-dinky Endrun Technologies Cf/Ct receivers are 1k each (GPS and
> CDMA variants)!  I was kinda shocked because there ain't much to these
> units AFAIK.  So I want to step back for a second (pun intended?) and
> ask everyone, WHAT SHOULD I BE LOOKING FOR?  Say I want nanosecond
> accuracy with respect to the PPS rising edge to absolute UTC....I know
> that for sure.  My target system is FreeBSD though I would like to use
> something with Snow Leopard as well (even if that's just ntpd synced
> locally to the FreeBSD box).  I am interested in running NanoBSD on a
> stand alone system.
>
> As for as assembly, I'm open, though I haven't soldered anything since
> undergraduate engineering school in a lab.  I am a software guy by
> trade.
>
> Another issue is what governs the price of these units?  I'd like to
> think its not going to be like the audio industry.
>
> Hope that helps,
>
> -aps
>
>    
Alexander

If you want the best performance possible from nanobsd then using an 
M12M or M12+T is advisable.
In this case hardware timestamping (with a resolution of about 100ns) of 
the PPS signal is used together with a customised version of the driver 
created by Poul-Henning Kamp.

The cheapest option would be to obtain used  M12+T timing receiver 
boards. however using these would entail constructing a board with a 
TTL/CMOS to RS232 transceiver and adding a 50 ohm driver for the PPS 
signal (required for highest time stamping precision as it avoids the 
delay of an RS232 driver and receiver). The only complication being that 
the receiver board uses a mating connector with 1.27mm pitch rather than 
the more common 2.54mm pitch connectors used by older receivers.

Of course network congestion and queueing delays add to the timing noise 
seen by client machines on the network.

If one used PTP (IEEE 1588) instead of ntp then the performance can be 
substantially improved if one can avoid standard network switches with 
their queueing delays.
However PTP is relatively new and as yet IEEE 1588gear hasn't yet hit 
the surplus market.
Usually PTP uses hardware timestamping of the relevant network packets, 
however there have been some efforts at software timestamping.

Bruce




More information about the time-nuts mailing list