[ntp:questions] NTP no internet connection

Richard B. Gilbert rgilbert88 at comcast.net
Sat Feb 9 21:50:29 UTC 2008


Unruh wrote:
> "Richard B. Gilbert" <rgilbert88 at comcast.net> writes:
> 
> 
>>flyersix wrote:
>>
>>>I have a network that I would like to cordinate the time in without
>>>exposing it to the internet for NTP servers.
>>>
<snip>
>>The problem with this is that your clock is not being disciplined.  This 
>>means that, in addition to not necessarily having the correct time, it 
>>may not be ticking at exactly 1 second per second.  Even worse, if the 
>>temperature is not controlled, the clock frequency and, therefore, the 
>>tick rate, can change from hour to hour.
> 
> 
> So what? The other clocks will simply follow that one, whatever its time.
> If its clock drift rate is 300ppm there might be trouble, but in general
> that is not what happens.
> 
> 
> 
>>Now, imagine other machines trying to synchronize with this server.  I 
>>think of it as one drunk driver trying to follow another.  Nobody is 
>>going to crash into anything but the various machines trying to 
>>synchronize to the server will probably be all over the map.
> 
> 
> What does "all over the map" mean. they will be withing a few tens of
> microseconds of that server. 

I mean that a system with a GPS based hardware reference clock 
synchronizes a mob of other machines more closely than a system using a 
bunch of internet servers can.  The later case is more of a "moving 
target".  Those internet servers MAY know the correct time to within a 
few microseconds but, by the time the internet is done with the query 
and response packets, the time is a "Scientific Wild Ass Guess" with an 
uncertainty of many milliseconds.  As ntpd changes its opinion as to who 
has the best time, and switches from server to server the time tends to 
bounce around a bit.  As the internet changes its behavior from 
night-time not very busy to day-time traffic jam the quality of the time 
deteriorates.  That "rock solid" GPS reference makes a big difference.





More information about the questions mailing list