[ntp:questions] NTP server with Garmin GPS 18 LVC

David Woolley david at ex.djwhome.demon.invalid
Fri Dec 20 09:42:04 UTC 2013

On 20/12/13 00:48, David Lord wrote:

> If ntpd is throwing away 7 out of 8 polls I'd say their results
> could be similar to mine.

Throwing away polls is only relevant when there is an abrupt change in 
local clock frequency.  In the steady state, the real situation is that 
ntpd is polling 8 times faster than it needs to given the low pass 
filter it is then going to apply.  In fact it is more than 8 times, as 
ntpd oversamples, even after you apply the minimum delay from last 8 filter.

The only way that ntpd could use those extra polls more effectively is 
to look to see if the offset was diverging consistently, indicating a 
clock frequency change.  The problem in doing that is how does it 
distinguish that from, say, a batch of high network delays with each 
delay longer than the previous one.

If it could, successfully, discriminate between the cases, it could 
change the filter bandwidth sooner than it actually does.

More information about the questions mailing list