[ntp:questions] Strange jumps in PPM

David Woolley david at ex.djwhome.demon.invalid
Thu Aug 22 21:24:45 UTC 2013

On 22/08/13 19:02, unruh wrote:

> ntpd assumes that the source is perfect. It has not choice. It does
> assume that network delays etc can make the measurements inaccurate, and

I was contrasting it with the fact that it is rather reluctant to accept 
that the local clock has wandered off, compared with assuming that there 
is wander on the measurements of the  upstream.

> tried to get rid of those by averaging, and filtering. (throwing away
> 87% of the emasurements) which makes its response slow. But it assumes

No.  They make it over-sample.  The poll rate is adjusted so that 1 in 8 
samples gives updates faster than needed for its estimate of the optimum 
loop bandwidth.

> that the sources are perfect without other evidence (eg disagreement
> between sources).

More information about the questions mailing list