[ntp:questions] Re: Root Dispersion and other parameter's effect

Hans Jørgen Jakobsen hjj at wheel.dk
Fri Nov 14 21:15:05 UTC 2003

On Thu, 13 Nov 2003 14:01:52 -0500, Bob Payment wrote:
> Folks,
> My customer asked me what the "maximum" values are for the NTP quality
> parameters that can be displayed on his Cisco routers.
> I believe that it "depends" but can anyone give some general guide
> lines as to what ballpark these values should be in if you wanted to
> maintain 1 msec, 5 msec or 10 msec time difference between two devices
> that sync back to the same master?
> Here is my customer's question.......
> Problem Description:  Hello
> Previous tac cases have shown how important good NTP synch is. However
> it is unclear what are the performance characteristics of NTP (or
> better said performance specifications) for support of SAA jitter
> measurements.  for example when one runs the cmd " Show NTP Status"
> what are the maximum values permitted for each of the four principle
> categories:
> clock offset 
> root delay 
> root dispersion 
> peer dispersion 
> I had NTP synch on my three test boxes but was still not able to see
> OW
> delay values. Again I had synch. When I switched to a different clock
> with
> improved NTP values (show NTP status) things worked much better. So
> why if I
> have synch (thus one assumes that all is hunky dory) do I not get good
> OW
> delay results with SAA. The answer is that the root and peer
> dispersion
> values were crap. OK so what are the thresholds one should obey with
> NTP to
> consider the NTP status good enough to support SAA jitter tests with a
> focus
> on OW delay values?

Think about what you are trying to do:
You are syncing the clock on the routers from remote clocks.
The packet for NTP travels on some loaded links (with asymmetric delay)
You are trying to make (one way) delay and jitter mesurements through
the same links.
One way measuments requre the that ntp is synced to local clocks or
at least to clocks that is reached through non congested links.

(A short answer could be: Dont sync the Cisco NTPs over the links
that SAA is making measurement on! )

Possible other points
1) What service class is used for the NTP packets?
   (request for feature: Be able to configure)
2) Cisco NTP seems to converge very fast to a low offset.
   Faster than ntp from Udel.
   Could there be mangled with the loop parameters in a way that would
   give priority to low offset.
3) A (Cisco) NTP using the burst options in the newer ntp implementations
   might help.

More information about the questions mailing list