[ntp:questions] stupid, simple question about precision

Richard B. Gilbert rgilbert88 at comcast.net
Tue Nov 21 01:46:52 UTC 2006


Simple Simon wrote:
> rgilbert88 at comcast.net wrote...
> 
>>David J Taylor wrote:
>>
>>>Simple Simon wrote:
>>>[]
>>>
>>>
>>>>My understanding is the best accuracy I might get, on a generic PC
>>>>clock, is about 0.050 ms--and that's with SSP and the kernel clock.
>>>
>>>
>>>My really old 133MHz Pentium PC with FreeBSD and a PPS source (with an 
>>>incomplete view of the sky) does a little better than than.  I'm sure more 
>>>modern hardware and a better-positioned GPS antenna would do substantially 
>>>better.
>>>
>>>  http://www.david-taylor.myby.co.uk/mrtg/pixie_ntp.html
>>>
>>>  http://www.david-taylor.myby.co.uk/ntp/FreeBSD-GPS-PPS.htm
>>>
>>>Cheers,
>>>David 
>>
>>I haven't tried it with a PC but a ~10 year old Sun Ultra 10 with a 440 
>>MHz processor and Solaris 8 keeps time within a couple of microseconds. 
>>   I do have a good view of the sky.  Something blazingly fast with a 
>>lower interrupt latency might do still better but my modest hardware 
>>meets and exceeds my very modest needs.
> 
> 
> Is your GPS directly updating your hardware, or is your SPARC an NTP 
> client of a Stratum 1 timeserver connected to the GPS, or... ?
> 

My GPS is connected directly to serial port "B" on the Sun Ultra 10. 
The PPS signal is connected to the DCD Pin.  The Sun also uses a bunch 
of Internet servers as a backup and sanity check.  Normally, the GPS is 
the selected synchronization source.  This gives me a platform from 
which I can evaluate internet servers and NTP over internet.

I find that servers/network connections are of highly variable quality. 
  I have ONE internet server that is solid as a rock and maintains an 
approximately 2 millisecond offset from my GPS.  I assume this is due to 
an asymmetry in the network somewhere.  A couple of the others bounce 
around +/- ten to fifteen milliseconds.  The quality of some of the 
others varies with the time of day; the hours of darkness seem to 
provide the best performance while normal business hours in the Eastern 
time zone (US) seem provide the worst.  I expect that this correlates 
strongly with network traffic.




More information about the questions mailing list