[ntp:questions] Query about NTP accuracy

Richard B. Gilbert rgilbert88 at comcast.net
Tue Jun 2 13:44:07 UTC 2009

Mike K Smith wrote:
> On 2 June, 14:31, Brian Utterback <brian.utterb... at sun.com> wrote:
>> David Woolley wrote:
>>> smithmi... at googlemail.com wrote:
>>>> In the case below, will server 3 be treated as a falseticker because
>>>> its offset lies outside the dispersion interval of the other two
>>>> servers even though there is some intersection between its dispersion
>>>> and that of server 2.
>>> That's basically why I won't be convinced by the four server argument
>>> until I've checked the actual code.  I seem to remember that there is an
>>> additional constraint, namely that the actual measurement must fall
>>> within the intersection.
>>>> 1
>>>>           |--------+--------|
>>>> 2
>>>>               |----------+----------|
>>>> 3                              |------------+------------|
>> It's the other way around. The interval is expanded to include the
>> mid-points. So, even though the overlap between server 2 and 3 is that
>> small section, the interval actually used must include their
>> mid-points (i.e. their offsets) so the interval used in the downstream
>> calculations is the offset of server 2 through the offset of server 3
> So the downstream servers reject server 1 as a falseticker as its
> offset falls outside that interval?
> They are then unable to select either of server 2 or server 3 as the
> synchronisation peer since the offset of each is at the extreme end of
> the intersection interval?

Perhaps there are good reasons for configuring five or seven servers?

A case for maintaining your very own atomic clock?

Or both?

More information about the questions mailing list