[ntp:questions] Re: NTP "maximum error" on local network

Richard B. Gilbert rgilbert88 at comcast.net
Thu May 18 17:44:12 UTC 2006

ajit.warrier at gmail.com wrote:

> The ntp server is not synchorinizing with remote internet servers, it
> is relying only on the local clock, so I do not think "clock-hopping"
> is the problem.
> We measure the 60ms delay as follows. A node sends a UDP packet from
> the server A to the client B. At the MAC layer on A, we put the
> timestamp of A in the packet. At the MAC layer at B, we put the time
> stamp at B. Finally at the application layer at B, we receive the
> packet and print the time stamps. These timestamps show a gap of almost
> 60ms, while NTP claims an offset of 4us. This is what confuses me.

See RFC 1305 p. 100 for an explanation of how offset and round trip 
delay are calculated.

Let T1 be the client timestamp in the request,
     T2 be the server's receive timestamp,
     T3 be the server's transmit timestamp, and
     T4 be the clien'ts receive timestamp.

Offset = ((T2-T1) + (T3-T4))/2
Delay = (T4-T1) - (T3-T2)

The above is taken from RFC 1305.  The RFC has a diagram and greek 
letters to further clarify the subject.

More information about the questions mailing list