| Re: [chrony-dev] handling of unstable sources |
[ Thread Index |
Date Index
| More chrony.tuxfamily.org/chrony-dev Archives
]
On Tue, Jul 13, 2010 at 02:10:15PM +0200, Frank Muzzulini wrote:
> Miroslav Lichvar wrote:
> >Ok, it's exactly as you said, 10.10.9.14 is almost always prefered
> >because it reports zero root dispersion. And when it makes the 5ms
> >correction it may be marked as jittery and 10.10.9.13 will be selected
> >as it's the only selectable source.
>
> So Mills' conservative calculations lost against the claimed
> precision of a stupid implementation. I admit that this is hard to
> cope with automatically.
>
> Maybe you can add a minimum root dispersion to be used when a source
> claims to be more precise than is reasonable.
How do you define reasonable? :)
The root dispersion value in NTP packet has only about 15us
resolution. Ntpd assumes a fixed frequency error and adds to the
dispersion 15 us every second since last reference update, but chrony
uses the estimated skew instead, so with a good reference clock it
will always transmit zero dispersion.
Anyway, I'd like to rework the source selection function to use also
some of the calculated statistics instead of just stratum and root
distance.
--
Miroslav Lichvar
---
To unsubscribe email chrony-dev-request@xxxxxxxxxxxxxxxxxxxx with "unsubscribe" in the subject.
For help email chrony-dev-request@xxxxxxxxxxxxxxxxxxxx with "help" in the subject.
Trouble? Email listmaster@xxxxxxxxxxxxxxxxxxxx.