2009/01/06 Allan Deviation Plot of local NTP server
As one may have noticed I'm a real time-nut. Not sure why but I want the time-keeping of my PCs as accurate as possible.
So of course I installed the NTP-daemon (NTP is a protocol for accurate time keeping on for example a computer) on all systems and even connected a DCF-77, MSF-receiver (DCF-77 and MSF are time-signals broadcasted from Germany and the UK) and a GPS (GPS satelites have a very accurate atomic clock on board) to some of them.
Now it seems one wants to verify if the NTP daemon behaves correctly. For that you can create an Allan Deviation Plot from the 'loopstats'-file. The Allan Deviation (a variation on the Allan Variance) then tells you how many PPM your clock drifts.
I googled a little bit and found this Octave (a free MathLab implementation) script. With some tweaking and adjusting I created a version which accepts 2 parameters (loopstats-file and PNG output-image filename) and a script which automatically graphs all your NTP loopstats-files.
An example result:
It is the evaluation of my DCF77 receiver. If I understand it correctly it means how many PPM the clock drifts (y-axis) for the time interval seen on the x-axis. A low Allan variance is a characteristic of a clock with good stability over the measured period. Here you can view some typical graphs for certain devices.
A tar-file including the altered .oct-file, the loopstats-graphing script as well as a script which can graph the peerstats files as well can be found here.