[ntp:questions] Re: What does refer to?

David Woolley david at djwhome.demon.co.uk
Fri Nov 11 07:32:03 UTC 2005

In article <pan.2005. at mail.localhost.invalid>,
"Tim <tim at mail.localhost.invalid>" <> wrote:

> script which is run after a PPP connection goes up.  But otherwise you end
> up with a computer no better than if you *never* had NTP on it.

That's not true.  The last available frequency correction is still applied.
There's no difference in this respect whether or not you have a local 
clock as a local clock never has a time offset so never causes any
adjustments; all it does is to make it look as though the time is
valid, so that you continue to serve time other machines.

If that's not happening, you have a bug, but you will need to install
an unpatched version, before reporting it here, or you could report
it to Red Hat.

If I remember correctly, if you are using the kernel PLL, this will still
maintain the frequency correction even after NTP stops in a controlled
way.  It will always continue to maintain it if NTP crashes.

> > When used as a fallback on servers, local clocks compromise the ability to
> > estimate the uncertainty in the time, because root dispersion and delay
> > are reset when the local clock is selected.
> I'll have to read more on that, because I don't understand the jargon.

NTP maintains the accumulated round trip time across all the servers
back to the reference clock.  This is called root delay, and half its
value sets an upper bound on the the time error due to assymmetry in
propagation delays.  It also maintains an estimate of the uncertainty
due to uncompensated drift, and various other factors, due to the time
delay between the reading of the reference clock and the current time.
This is root dispersion.  If you use the local clock, root delay is
zero and root dispersion is only the local part of the other factors,
and doesn't increase with time.

More information about the questions mailing list