PPS over USB

[Dan Drown]

Quoting "Eric S. Raymond" <esr at thyrsus.com>:
> Dan Drown <dan-ntp at drown.org>:
>> >Eric: Please plug your 701W into your Pi with HAT and collect some data.
>>
>> Here's 7 days worth of data:
>>
>> https://dan.drown.org/rpi/usb-vs-gpio-pps.html
>>
>> The Rapsberry Pi 2's GPIO-PPS based local clock offsets at 0.1 percentile
>> and 99.9 percentile were -3.61 microseconds and 4.13 microseconds.
>>
>> The Raspberry Pi's USB-PPS based local clock offsets at 1 percentile and 99
>> percentile were -582.97 microseconds and 539.08 microseconds.
>
> Dan and Hal, how do you interpret those results? I would naively suppose
> that the listed fudge should be the mean of each distribution (assuming
> it looks Gaussian) and thius the arithmetic average of those pars
> - am I missing anything?

The fudge setting is a different thing altogether.  This is more along  
the lines of the level of precision available.

I'm using this as my fudge (I think I'm getting the trailing edge of a  
100ms pulse):

fudge 127.127.22.1 refid PPSU time1 0.100700

I use the 50 percentile offset of my nearby clocks to set the fudge factor.

One example graph between the Pi2 and a third machine to show the 50  
percentile value:  
https://dan.drown.org/rpi/remote-statistics.10.1.2.19.png