[ntp:questions] Re: NTP Polling Explanation

David Woolley david at djwhome.demon.co.uk
Tue Aug 15 21:26:14 UTC 2006


In article <49390eba0608150658o16896df0ra6eb1f98812b1b42 at mail.gmail.com>,
whmyers at gmail.com (Bill Myers) wrote:

> local system clock.  Shorter polling intervals cause NTP to make large but
> less accurate calculations that never stabilize, causing the local system
> clock to wander.  Longer polling intervals allow NTP to calculate smaller

It would take some time to read the code and spec closely enough to be
completely sure, but I believe that maxpoll doesn't actually cap the 
loop time constant, so the loop is still less responsive to jitter.  However,
by setting maxpoll you are wasting resources by oversampling at a rate that
is simply filtered out by the low pass effect of the control loop.  Also,
ntpd chooses the sample from the last 8 that has the tightest error bounds,
where the bounds increase with time as well as depending on things like
round trip time.  If there is an 8 minute period with extended round trip
times because of an assymetric traffic load, maxpoll=4 will cause an
ureliable sample to be used, but the default polling interval will reject
that period.

As I understand, if the poll rate is faster than required, the sensitivity
will be reduced to compensate.

I think the one advantage of limiting maxpoll is that the system will detect
a step-out (>128ms) condition faster, but its better if the system doesn't
get those in the first place.




More information about the questions mailing list