[ntp:questions] ntpd step recovery time

unruh unruh at invalid.ca
Tue Apr 24 01:01:37 UTC 2012

On 2012-04-23, Chris Albertson <albertson.chris at gmail.com> wrote:
> On Mon, Apr 23, 2012 at 1:54 PM, David Woolley <
> david at ex.djwhome.demon.invalid> wrote:
>> You are trying to something as a normal operation that is actually a last
>> resort error recovery strategy.  The specification writer doesn't
>> understand the purpose of ntpd.
> I've seen this happen before, a customer does a cut and paste out of some
> specification and they really don't know what they pasted into the
> document.  And then your management is afraid to insult the customer by
> pointing out their error.      Best you can do is correct the problem in
> your test plan by interpreting the requirement in some way that actually
> makes sense.    But really this is such a grossly mis-written requirement
> that you should go back to the author.

Not clear. Yes, the step error is highly unlikely but not impossible.
How well does ntpd recover if that happens. The OP complained that the
response time was too slow. One answer is to limit the poll interval.
Thus instead of poll 10, limit the poll to poll level 6 or 7 say. Yes,
youshould then make sure that you are using your own ntpd server so as
not to overload some network server. An even better idea might be to put
a gps onto the computer with poll interval 4. 
It is certainly true that ntpd cannot react very quickly when running at
poll 10.
It does come at the cost that if the access to the servers disappears
entirely, the rate of the clock will have a larger error than it would
have if you used a poll 10, so you have to balance rapidity of response
to rare jumps, with stability in case all servers disappear. 

> Chris Albertson
> Redondo Beach, California

More information about the questions mailing list