[ntp:questions] Jitter versus polling interval
unruh-spam at physics.ubc.ca
Sun Mar 1 18:43:30 UTC 2009
"David J Taylor" <david-taylor at blueyonder.neither-this-bit.nor-this.co.uk> writes:
>David Woolley wrote:
>> David J Taylor wrote:
>>> Does anyone know an approximate relationship between polling interval
>>> and peak-to-peak jitter for a non-ref-clock system? I.e would the
>>> jitter offset) go up linearly with polling interval, or as the
>>> square, or what?
>> Do you mean true jitter (RMS error from true time) or the jitter
>> statistic produced by NTP?
>Neither, directly. I mean the peak-to-peak variation in the reported
>offset, as "measured" by eye on a graph like the "Daily" graph here:
>> I may be completely wrong, but I thought that true jitter initially
>> went down with increasing polling interval and that ntpd stopped
>> increasing the polling interval before it started to go back up
>> again. That's because ntpd increases the loop time constant as it
>> increases, which compensates for the increased poll interval and, in
>> the Millsian world, measurement jitter has a 1/f spectrum.
>> In the real world, you'll need to specify the nature of the
>> measurement noise.
>Yes, I'm looking for an approximate value only at the moment - an order of
>magnitude relationship to say whether its linear or quadratic (with number
>of seconds). I'm thinking of a system where the maximum polling interval
>might be set to 7, 8, 9 or 10.
YOu have a couple of answers, both of which are probably wrong and are
asking which it is. As stated your premise is bad. It depends is the best
answer you are going to get. It depends on the source of the jitter.
Measurement noise? clock drift noise? assymetric network delays? Day of
week? phase of moon?..... All make a difference.
And it depends on how it is compensated. Mill's Markovian feedback?
Chrony's linear regression? ...
More information about the questions