[ntp:questions] Poll timing

Richard B. Gilbert rgilbert88 at comcast.net
Thu Aug 16 21:02:53 UTC 2007


Steve Kostecke wrote:
> On 2007-08-16, Jussi Kauppinen <jtkauppi at cc.hut.fi> wrote:
> 
> 
>>Paul.Croome at softwareag.com wrote:
>>
>>
>>>Why should the exact instant when NTP polls be important? It doesn't
>>>matter whether the polling interval is 64 seconds or 64.1 seconds or
>>>whatever.
>>
>>I'm developing a system that utilize NTP. Unfortunately, the details of 
>>the system are confidential. For this system to work properly, it is 
>>essential that polling phase inside a second is approximately constant 
>>related to the clock of the client,
> 
> 
> The exact poll timing is irrelevant as far as NTP is concerned.
> 
> The Unicast (i.e. client/server) polling scheme time-stamps the packet
> when it is sent from the "client", at the "server", and again when it
> arrives back at the "client". These three time stamps are collected at
> each poll interval from each time source. The NTP algorithms use these
> time-stamps to, over time, determine what the correct time is.
> 

I believe that the packet is timestamped TWICE by the server; once when 
it arrives and again when the response is returned.  These two would 
normally be very close to each other.  See, for example, page 100 of 
RFC-1305.







More information about the questions mailing list