Pivoting

[Achim Gratz]

> Huh?  Potentially you need to apply a pivot to every packet that comes in;
> different sources could even have different epochs.

Nope.  All calculations are done in modulo arithmetic and end up as time 
deltas to local time.  One of the assumptions of a working NTP is that 
the local times of all systems involved do converge to true time, so 
those differences are _much_ smaller than 68 years around local time.

> True.  But it is not clear what we can do about this other than what we are
> already doing - that is, pull in lots of sources, discard outliers, make
> a best guess.

At the minimum we'd need to document the assumptions.  As said before, 
the more independent estimates of time (or rather bounds on the time) 
you have, the less likely it becomes to guess wrong on the NTP era at 
startup.

-- 
Achim.

(on the road :-)