[ntp:questions] On NTP server, how to check offset between client and NTP server?

David Woolley david at ex.djwhome.demon.co.uk.invalid
Thu Oct 9 22:16:36 UTC 2008


chowalit.lab Chowalit Lab Linux wrote:
>       I have stratum 2 NTP server. My client request time from my NTP
> server. On my NTP server, how to know offset of between my client and
> my server ? I found ntp_script software suite from
> http://www.schlitt.net/scripts/ntp/. It's not solve my problem
> directly.

When NTP is working properly, the difference between the time on the 
client is zero within statistical error.  The "offset" will, more often 
than not, exceed the statistical error in the actual time difference.

There are situations, in which this doesn't hold true for NTP, 
particularly within a few hours of system startup (and also after sudden 
temperature changes), and these are discussed in the thread entitled 
"Slow convergence of NTP with GPS/PPS", although the situation will be 
more extreme there, as one can read a reference clock with less 
statistical error than an ethernet connected server.  You can get more 
on this theme from most of the recent long running threads.

Once you've read and understand these, if you still find you want to 
measure the figure that ntpd calls "offset", you treat the client as a 
server and run ntpq on the real server, and use the peers command.

If you want to measure the actual offset between the clocks, you will 
need to add special instrumentation to the kernel, and special hardware. 
  You will then want to ask yourself why you don't use the error 
measured by that instrumentation to apply corrections to reduce that 
error to the statistical noise level!

But, again, remember that any time synchronisation protocol optimised 
for minimum phase error will never be able to measure the actual error, 
as any error it can measure it will have corrected.




More information about the questions mailing list