Re: [gpsd-dev] Clarifications needed for the time-service HOWTO

[Andy Walls]

On Mon, 2013-10-21 at 08:00 -0400, Eric S. Raymond wrote:
> Hal Murray <address@hidden>:
> > 
> > address@hidden said:

> TODO: more on performance tuning and diagnostics here. 
> 
> This is where we need to talk about (a) adusting fudges, and (b) how
> to check the quality of your chimers.

WRT adjusting fudges:

http://lists.ntp.org/pipermail/questions/2011-November/030951.html

I happen to use the "calibrate by hand" method mentioned in the post.
Open-loop is easier for me to follow.

I also have this cheat-sheet lying around for examining the data from
open-loop tests:

        http://www.eecis.udel.edu/~mills/ntp/html/monopt.html

        loopstats:
        clock offset (s)                offset
        freq offset (PPM)               drift compensation
        RMS jitter (s)                  estimated error
        RMS freq jitter (PPM)           stability (aka wander)
        clock poll constant (log2(s))   polling interval
        
        peerstats:
        clock offset (s)                offset
        roundtrip delay (s)             delay
        dispersion (s)                  dispersion
        RMS jitter (s)                  skew (variance)
        
        Show the PPS deviation from system clock. (Keep in mind that PPS
        is probably better than the NTP adjusted system clock.)
        
        gnuplot> plot    "peerstats.pps" using ($2):($5) with lines 1
        gnuplot> replot  "peerstats.pps" using ($2):($5):($8) with yerrorbars 2
        
        Show the GPSD shared memory clock deviations from the system
        clock.  (You will get the GPSD shared memory clock fudge value
        estimate from this data when NTP has converged to your
        satisfaction.)
        
        gnuplot> plot    "peerstats.shm" using ($2):($5) with lines 1
        gnuplot> replot  "peerstats.shm" using ($2):($5):($8) with yerrorbars 2
        
        Check the tracking loop convergence (phase):
        
        gnuplot> plot    "loopstats" using ($2):($3) with lines 1
        gnuplot> replot  "loopstats" using ($2):($3):($5) with yerrorbars 2
        
        Check the tracking loop convergence (frequency):
        
        gnuplot> plot    "loopstats" using ($2):($4) with lines 1
        gnuplot> replot  "loopstats" using ($2):($4):($6) with yerrorbars 2

Note that I had run some grep commands to separate PPS peerstats from
SHM peerstats.  Note that I also do not feed PPS to NTP via GPSD, but
via a /dev/pps device unrelated to the serial port of the GPS (PPS is
wired to a GPIO).

For estimating GPIO connected PPS fudge, an experiment in measuring GPIO
IRQ latency is needed (something like this:
https://github.com/scottellis/overo-irqlat )


Regards,
Andy