[ntp:questions] strange behaviour of ntp peerstats entries.

Unruh unruh-spam at physics.ubc.ca
Sat Feb 2 14:06:42 UTC 2008


"David L. Mills" <mills at udel.edu> writes:

>Guys,

>This is really silly. The Unruh agenda is clear. Should you choose to 
>limit the application space to fast local networks, the chrony choice 
>may or may not be optimal. Should you extend this space to the raunchy 
>global Internet, conviction will require diligent testing and analysis. 
>There is no clear evidence the chrony algorithms are sufficiently agile 
>to osicllator wander over the long term and I see no response to this 
>issue.


If you thing the Unruh agenda is clear perhaps you could tell Unruh (me)
what it is you think it is. Again you seem to be resorting to innuendo as
an arguing tool. I thought my agenda was to improve ntp by
pointing out that there is an alternative algorithm which works better
than ntp does on a real network. The netroks I have available are limited,
so to ask if it is really better on the rauncy global internat is a fair
questions. But to imply that it is not sufficiently agile to handle
oscillator wader is simply wrong. It is. It is at least as "agile " at ntp
and in fact is much more agile--- it repondes to changes inthe oscilator
frequency much faster than does ntp, and as reliably. I have shown you data
on the web page where chrony responded to oscillator wander far bettter
than does ntp.

Whether or not it responds to bad networks, highly variabale round trip ,
etc better than ntp I do not know and you are right it needs more testing.
I have tried to alert the ntp community to a potentially better way of
doing things. Your attitude may be that you are going to close your eyes
and ears unless it is proven better beyond any shadow of a doubt.Or you
might study it to see what could be learned. It IS different from ntp. My
tests show that at least on the testbed I have, it works better than ntp,
significantly better in terms of the offset variance, (maybe because ntp
throws away 80% of all measurements, maybe bacuse of its long time
constant, maybe for other reasons). It may be that it does not do so for
the more raunchy networks. 
I am now convinced it is not a feedback loop. It estimates the slope and
drift from the immediate past (10-100 samples) It cannot go unstable unless
it is through an interaction with  measuerment error(I have zero evide3nce
of that but am keeping my mind open to the possibility).

That was what I thought my agenda was. 

>Dave

>David Woolley wrote:

>> Steve Kostecke wrote:
>> 
>>> There's nothing stopping him from implementing what he considers to be a
>>> solution himself. He could even distribute his modified version of NTP
>>> to anyone who wanted to use it.
>> 
>> 
>> Why should he do that when something already exists, although it is not 
>> technically NTP?  As I see it, he is trying benefit ntpd by encouraging 
>> it to behave as well as the alternative that he is using.
>> 
>> There are two ways that people can vote by actions:
>> 
>> 1) to implement the features in the (open source) original;
>> 2) to switch to an alternative that already has the behavour they 
>> consider desirable.
>> 
>> You cannot say that the criticism is invalid because people choose 
>> op;tion (2).
>> 
>> There is also a number (3), which is to be aware that there is a problem 
>> but be prepared to live with it.  I'm probably in that category.
>> 
>> There is probably also 1a, which is what resulted in the alternative, 
>> which is to implement a competitor, more or less from scratch, which 
>> only has the features one considers important.




More information about the questions mailing list