[ntp:questions] time to sync vs ptp?

[David Woolley]

matthew.garman at gmail.com wrote:

> 
> OK.  But at some point, when the drift file is established, it has a memory of sorts, right?

Not in the sense that unruh means.  If you ignore the minimum delay 
filter, ntpd uses an IIR filter with only z^0 and z^1 terms, so it only 
remembers the state from one accepted sample back.  Linear regression 
based system actively use values from several back samples in each 
iteration.

The drift file provides a checkpoint of part of the state, but is only 
used after an outage.

You can consider the minimum delay filter as using z^0 to z^7, but it 
ends up only selecting one of these, even though it uses all of them to 
make the decision.