[ntp:questions] Re: ref clock for network POP locations

Richard B. Gilbert rgilbert88 at comcast.net
Tue Sep 26 20:45:23 UTC 2006


Jeff W. Boote wrote:

> I have some network locations where I have no way to access GPS or CDMA. 
> But, I would still like to have a very accurate time source. (I would 
> like sub-microsecond accuracy from NTPD if I can get it.) These 
> locations are in telco-hotels, so they are typically very well shielded 
> making any broadcast technology unworkable. (We are not using sonet - so 
> getting time from there is also not an option.)
> 
> I was thinking that it might be possible to get an oscillator to provide 
> a PPS signal, and then use a very well connected (possibly even a 
> dedicated network link with no router/switch) network peer as the 
> preferred peer. Is this a reasonable solution? Can anyone suggest 
> anything better?
> 
> The PPS would need to be calibrated... Is the normal ntp calibration 
> reasonable for this? I would guess it depends on the oscillator...
> 
> If this is a reasonable solution, does anyone have recommendations for 
> oscillators? Most of the ones I have seen want a reference source of 
> their own. This of course does not work in my case...
> 
> Thanks,
> jeff
> 
> P.S. I will happily add information to the wiki on this after figuring 
> out how to do it. (If?)

The only ways I know of to get "sub microsecond accuracy" from NTPD are:
a. a GPS based reference clock (the PPS output is usually rated at 25-50 
nanoseconds)  By the time you get to the computer's clock that degrades 
to a microsecond or two.
b. a CDMA timing receiver
c. a LORAN receiver
d. an atomic (Cesium) clock

Forget about doing it over a network.  It might be possible but I 
suspect that a Cesium frequency standard with clock option would be 
easier and cheaper. (I believe that Cesium clocks go for $40,000 and up!)

The problem with networks is that they introduce random "phase noise"; 
e.g. if you set up a server in New York that sends packets at EXACT one 
second intervals, those packets will not arrive in Los Angeles at 
anything like one second intervals.  Some of them may not arrive at all 
if you are using UDP.  The best you can hope for using internet servers 
is 2-5 millisecond accuracy.  The worst is much worse than that; much 
depends on the quality of the servers involved and more depends on the 
quality of the network connection between you and the server.

If worst comes to worst, you can put the GPS receiver on the roof and 
run fiber from it to where you need the time.

FWIW, Telcos tend to have some pretty serious timing stuff of their own; 
   Cellular phones, T1 and T3 lines all rely on very precise timing. 
This may take the form of frequency standards rather than time standards 
but the two are intimately related; t=1/f and f=1/t




More information about the questions mailing list