[ntp:questions] Re: Excessive polling

David L. Mills mills at udel.edu
Tue Dec 2 16:45:16 UTC 2003


Tim,

As I said on the hackers list, when a reference clock is reachable the
poll was clamped at the reference clock poll, which normally is 64 s. I
changed that so it is no clamped at the discipline loop poll, which
ordinarily creeps up to maxpoll. While reference clocks themselves are
still oversampled at 64 s, the other critters now follow the loop poll.
All this had a surprising (to me) side effect of improving the reference
clock timekeeping to the order of a couple of tens of microseconds, even
without PPS. See pogo.udel.edu. Darn.

Dave

Tim Shoppa wrote:
> 
> "David L. Mills" <mills at udel.edu> wrote in message news:<3FBA502A.C9127E29 at udel.edu>...
> > NTPv3 (xntpd) operates as you describe; the poll intervals are clamped
> > to the reference clock. NTPv4 (ntpd) has no such restriction; the poll
> > intervals cheerfully escalate for sources other than the reference
> > clock.
> 
> That's not what I observed.  I'm running 4.2.0 on linux, and the
> reference clock (I use refclock_hpgps on one of my stratum-1's and
> refclock_wwv on my other) is polled every 64 seconds, and all the servers
> are polled every 64 seconds too.  This is true even after weeks of
> stable running.  If I remove the refclock from ntp.conf
> then the poll interval to the network servers go to 1024 seconds very quickly.
> 
> In order to be a "nice" network neighbor, I explicitly pegged minpoll
> and maxpoll so that they were both at 1024 seconds.
> 
> I'm not accusing the 4.2.0 maintainers of explicitly doing this, in fact
> I read the source code and it looks to me like what you say is true.  But
> in operation that's not how it works.
> 
> Tim.



More information about the questions mailing list