[ntp:questions] how will this effect stability

David L. Mills mills at udel.edu
Fri Feb 16 18:10:56 UTC 2007


Uwe,

While USNO is committed to Kalman filters, NIST is committed to ARIMA 
methods. Both of these methods are quasi linear and don't work well in 
the messy Internet. The ntpd method uses seriously crafter nonlinear 
amputation of delay spikes followed by nonlinear selectio and averaging 
algorithms. Think of this approace as a noise limiter and bandpass 
filter used on high-end communications receivers.

Dave

Uwe Klein wrote:

> Terje Mathisen wrote:
> 
>> Uwe Klein wrote:
>>
>>> Terje Mathisen wrote:
>>>
>>>> Please let huffpuff run for a few days while plotting frequency and 
>>>> offset values, my guess is it will turn out to be darn stable. :-)
>>>>
>>>> Terje
>>>>
>>> some data to gloat over:
>>> huffpuff is at 160000 added ~19:00 on the 14th
>>> t-online was added later.
>>>
>>> t-online.png is the ntp server of my provicer ntp1.t-online.de over dsl
>>> ptb.png is the same to ptbtime1.ptb.de , dsl and then some more hops
>>> robert.png is to another host on the local network.10Base100
>>>
>>> http:// home.foni.net/ ~kuttersegeln/ntp/
>>
>>
>>
>> So, are you happy with those results?
> 
> I am holding my breath.
> 
> ( one thing I noticed this morning is that ntpd hops between
>   the two Stratum 1 servers )
> 
> ( wouldn't a Kalman Filter Type approach work better with
>   time sources of comparable quality? )
> 
> enabled more logging and wrote some scripts to collect and plot data.
> going to post some more images tonight.
> 
> One thing I noticed is that logfile statements position in the config
> file is significant.
> i.e. when logfile stuff ist last in the config the initialisation
> logging goes to default ( syslog )
> Does the same apply to things like the "tinker" stuff?
> 
> 
>>
>> You will always see horrible delay/offset values, but the local clock 
>> should coast over those periods, keeping the local frequency pretty 
>> stable.
>>
>> In fact, plotting the clock frequency numbers alongside the other 
>> values would be a very nice way to determine if/when things have 
>> stabilized.
>>
>> You should see either a straight line, or a sawtooth/sinus curve which 
>> corresponds to the air condition in your server room.
> 
> I kept the heating on this night, at its permanent location later on temp
> ist kept ~ constant ( central heating, no climate control ).
> 
>>
>> Terje
> 
> G!&Tks
> uwe
> 
>>




More information about the questions mailing list