[ntp:questions] Loopstats updated less frequently than expected

Unruh unruh-spam at physics.ubc.ca
Sat May 23 19:18:50 UTC 2009


"David J Taylor" <david-taylor at blueyonder.not-this-part.nor-this.co.uk.invalid> writes:

>Unruh wrote:
>[]
>>> As I have maxpoll set to 6, I would have expected entries no more
>>> than 64s apart.  What am I failing to understand here?  The poll
>>> values reported in ntpq -p are 64, 64 and 1024 - as expected.
>>
>> You are forgetting that ntp throws away 7/8 of the stuff it collects.
>> (This is in an attempt to counter round trip biases).
>> Thus you would expect to have 8 times the max poll interval between
>> loopstats. (it is random so it is not always exactly 8 times sometimes
>> less, but never more, but it averages out to pretty close to 8 times.)

>Thanks for that, Bill.  With the servers with a GPS directly attached, I 
>typically get 5400 lines in the log file, for a 16s polling interval. 
>That's one line per poll, not one line per approximately eight polls.

If you mean you have a refclock attached, they have round trip time of
0. Thus there is never a minimum and all get used. Ntp's algorithm in
short is: Save the last 8 polls. Check if the current poll has a larger
roundtrip time than any of those 8. If it does, do not use it. 
If the round trip times are all 0, then the current one is never larger
so it is always used. 

Now, the round trip time on the net  is a random variable, but from experience, on
average only about 1 in 8 are used. ( actually it is a bit better than
that so that about 1/7  are actually used.)

This is to avert bias on the theory that longer round trip times are
probably biased (  the delay is caused by one of the legs being longer than the other),
 but as a result, the actual average poll interval is
roughly 7 times longer than the requested interval. Also since the time
constants must be longer than the longest poll interval, this helps the system
to have a very long time constant (ie respond extremely slowly to
changes). 





More information about the questions mailing list