[You sent that to 3 lists that I'm on. I'll reply here. I think your X axis is off. Your left edge is 1 second and you don't have data for that. peerstats is every 64 seconds. (unless you mucked with maxpoll) I'm guessing the software you used is assuming they are every second. There may be a command line parameter. The normal ADEV is V shaped. The left slope is noise in the readings. The right slope is drift in the clock. That assumes you have a good reference clock, where good means much better than the DUT. In this case, there is no long term drift since ntpd is tracking GPS. You will probably get a traditional V if you disable ntpd. That will turn things inside out. Your DUT will be good (GPS) and your reference clock (PC) will drift. There is probably a simple way to do that but I can't think of it. I would try adding noselect to all the servers and refclocks, and add a localclock if you want to monitor that box from other systems. There is a slight step in the NMEA data at 1000 seconds and maybe a similar step in the PPS data at the right edge of the graph. I don't know what that means. -- These are my opinions. I hate spam.

Yo Hal! On Tue, 19 Jul 2016 14:28:32 -0700 Hal Murray <hmurray@megapathdsl.net> wrote: > [You sent that to 3 lists that I'm on. I'll reply here. Yeah, but time-nuts is the one list that knows more about this than I do. So good choice. > I think your X axis is off. Your left edge is 1 second and you don't > have data for that. peerstats is every 64 seconds. (unless you > mucked with maxpoll) I'm guessing the software you used is assuming > they are every second. There may be a command line parameter. I'm taking the time and offset from the /var/log/ntpstats/peerstats file. So the data is collected every minpoll (16 seconds), timestamped in seconds with millisecond precision. The offsets are also in seconds. Should I set the left edge to 16 seconds? > The normal ADEV is V shaped. The left slope is noise in the > readings. The right slope is drift in the clock. That assumes you > have a good reference clock, where good means much better than the > DUT. Well, the data is the data. I used Tom Van Baak's adev5 program to do the calculations. Given that the system is GPS stabilized on PPS I can't see why the accuracy should not get better over long durations. > In this case, there is no long term drift since ntpd is tracking > GPS. You will probably get a traditional V if you disable ntpd. Well, I'm trying to measure ntpd, so chicken and egg here. > There is a slight step in the NMEA data at 1000 seconds and maybe a > similar step in the PPS data at the right edge of the graph. I don't > know what that means. At Tom's suggestion, I upped the sampling from Octaves to 50/decade. I also added TDEV. Now there is some structure in there. https://rellim.com/graphs/adev.png Some weird things in there... RGDS GARY --------------------------------------------------------------------------- Gary E. Miller Rellim 109 NW Wilmington Ave., Suite E, Bend, OR 97703 gem@rellim.com Tel:+1 541 382 8588