[ntp:questions] refclock on Windows graphs

Dave Hart davehart at gmail.com
Tue Mar 17 02:04:54 UTC 2009


Unusually for me, the configuration and code on my reference clock
test machine has been static for the last two days UTC.  I've posted
loopstats and graphs for that period on http;//davehart.net/ntp/
refclock/

Most of all, the low jitter averaging under 3us suggests I don't have
a lot of room left for improvement before running up against the
claimed +/-1us accuracy of the GPS 18x LVC's PPS:

http://davehart.net/ntp/refclock/refclock-20090315-20090316-jitter.jpg
http://tinyurl.com/c4juho

To understand the two spikes in jitter look at either:

http://davehart.net/ntp/refclock/refclock-20090315-20090316-combined.jpg
http://tinyurl.com/dg7gsq

(combining frequency and offset) or separately:

http://davehart.net/ntp/refclock/refclock-20090315-20090316-freq.jpg
http://tinyurl.com/c85z4r
http://davehart.net/ntp/refclock/refclock-20090315-20090316-offset.jpg
http://tinyurl.com/dybvf3

The slow positive climb in frequency most of the time corresponds to
the temperature slowly declining.  There are a few transients pulling
the frequency up briefly, probably caused by a door being opened
letting in cold air.  The dramatic frequency drop in the middle is
caused by a furnace cycle, bigger than most suggesting the thermostat
was turned up.  I don't have temperature monitoring but it's on the
2nd floor with forced air heat controlled by a single thermostat on
the 1st floor.  The temperature upstairs continues to rise for some
time after the thermostat turns off the heat.

I wish this was completely representative of the new interpolation
scheme in my recent releases.  There is an unexplained jitter
variability between runs of ntpd.  Some runs it's several times
higher, but still an order of magnitude better than the previous
interpolation approach.  I'm expecting but have not yet seen
demonstrated that the new scheme works better on machines with 10ms
clock periods (all mine have 15.6ms clock periods).

Cheers,
Dave Hart




More information about the questions mailing list