[time-nuts] Divide by five ->Ensemble

Hal Murray hmurray at megapathdsl.net
Sun Nov 9 18:07:06 EST 2014

kb8tq at n1k.org said:
> What is going on is that people are confusing the estimation process that is
> used by the selection process (which does look at a lot of stuff) and how
> that is described.   ...

In this context, it's important to remember that there are 2 parameters 
associated with the goodness of clocks.  The first is the error.  The second 
is the error bars.  For NTP, the error bars are often huge and often differ 
wildly between clocks.

I don't know how that translates into getting the best out of a batch of Rbs.

NTP has another trick worth adding to the collection.  For things like PPS 
processing, it collects a batch of samples, then discards roughly 1/3 of them 
as outliers.  That helps keep an occasional bad sample from biasing the 

These are my opinions.  I hate spam.

More information about the time-nuts mailing list