[ntp:questions] Force ntpd to immediately trust server

David Woolley david at ex.djwhome.demon.co.uk.invalid
Sat Feb 16 10:37:47 UTC 2008


Harlan Stenn wrote:
>>>> In article <64B9BEB4F544624B9D59DB6F61E2E65402AA7C50 at AZ25EXM03.gddsi.com>, John.Johnson at gdc4s.com (Johnson, John-P63914) writes:
> 
> John-P> Can ntpd be configured so that on start up it **immediately** trusts
> John-P> a time source (remote server or local clock) and begins serving

You really need to read most of the recent postings on this newsgroup, 
as most relate to the start up behaviour of ntpd, and its slowness. 
There is a lot of controversy and some possible alternatives to ntpd, 
but I don't think there is any real consensus yet, so you need to read 
the discussion and make up your own mind.

> John-P> time?  For my application ntpd must be up and serving time in a
> John-P> matter of seconds.

Because ntpd cannot control its clients, it has only serve time that all 
clients can rely on, not just those with some, possibly, relaxed 
requirements.

> 
> How many seconds?  In many cases, we can do this in 11 seconds (and there is
> talk now about residual slew time).

The caveat here is that the time may only be good to 128ms, as ntpd will 
not step the time on startup unless (in the default configuration) it 
requires a change of more than 128ms.

Also, in rare, but not, I believe, impossible, circumstances, it may, 
very quickly, declare the time as invalid and initiate a recovery.  This 
would happen if the initial time offset reading was less than 128ms, but 
subsequent readings were just over 128ms.
> 
This whole area is fairly unstable, as there have been many changes even 
in versions which are all 4.2.x.  In particular, the 4.2.0 startup 
sequence is rather different from the 4.2.4p4 one, where the offset is 
more than 128ms, but the diftfile is present.




More information about the questions mailing list