[ntp:questions] Re: Measure clock accuracy, Store to a text file, Expected Oncore M12T accuracy?

Richard B. Gilbert rgilbert88 at comcast.net
Fri May 26 21:03:49 UTC 2006


Mario Behn wrote:
> Hi,
> 
> I want to measure how well my PC time is synchronized to UTC (e.g. true
> time). I have a oncore as a reference clock on an embedded system, which
> has now graphic libraries installed, so I won't use MRGT or so as I found
> as one suggestion.
> 
> So I would like to write out the estimated accuracy and the
> estimated standard deviation (= jitter?) to a simple ascii file, which I
> then can download and plot with matlab or so. 
> 
> Now I few questions:
> 
> 1.) Should I include more then the estimated accuracy and the
> estimated standard deviation? What about Allan variance (by the way what
> is that is the importance of this parameter) ?
> 
> 2.) What is the easiest way to perform this measurement? I know there are
> many files like /var/log/ntpstats/clockstats, which seem to already
> include some of this data? However I don't know whether thats the way to
> go?
> 
> 3.) Well finally after have got the measurement, I need to interpret it
> and decide whether it is reasonable for my system. Now I have a VIA EPIA
> MII 6000 with 600 Mhz C3 processor and running Red Hat 9.0 with a 2.4.32
> kernel and the corresponding pps kit. I have the synergy system smart
> antenna based on the oncore M12T receiver. What accuracy can I expect?
> 
> So thats for now, any ideas and comments are highly appreciated.
> 
> Thanks -- Mario
> 

I use gawk and a script to pull data from my peerstats file and plot it 
with Gnuplot.  Don't know if you have peerstats with your embedded 
system though.  Peerstats has all the servers jumbled together so the 
gawk script pulls data for each server and writes it to a temporary 
file.  Gnuplot extracts the requested fields from the temporary file and 
plots.

Interesting plots:
Offset vs. time of day
Delay vs. time of day
Delay vs. offset (scatter gram)
Synchronization distance (dispersion + delay/2) vs. time of day.

Perhaps the most interesting thing to study is how the internet can 
mangle what is presumably good data to start with.  I'm sure that most 
of the stratum 1 and stratum 2 servers are within a few hundred 
microseconds of the correct time but you couldn't prove it by what 
arrives here!




More information about the questions mailing list