David J Taylor wrote:
So, if I understand you correctly, you are saying that because ntp
measures and reports an offset, it will try and correct the clock in the
opposite direction, so that the actual clock error is likely to be less?
So will ntpdate, but ntpdate only uses the most recent offset, whereas
ntpd uses (to a first approximation) an exponential rolling average of
the recent offsets, such that it makes effective use of quite a lot of
them. That has a similar effect to using longer and longer averaging
periods in MRTG. The time of the clock is much more accurate than any
one offset would suggest.
Complicating factors are that there is a non-linear finite impulse
response filter (best of 8) ahead of the main, infinite impulse response
one, the actual filter is not just solving for an offset, but also for
the frequency, and the filter parameters are varied depending on an
estimate of the quality of the time data.
_______________________________________________
questions mailing list
[email protected]
http://lists.ntp.org/listinfo/questions
