[time-nuts] Histogram of T2 arrival times

[Hal Murray]

gem at rellim.com said:
>> The systems collecting the data have 200-300 microsec peak-peak of
>> clock offset, mostly tracking daily temperature swings.
> How do you know that? 

Which "that"?

I don't have a formal proof, just a collection of data that all fits together.

You can get the clock offset from loopstats.  That can lie, for example by using a PPS over USB without fudging to correct for the USB polling delay or getting the time from the net with asymmetric delays due to routing or last-mile link speeds.

You can confirm things by using a good NTP system to monitor your targets.  That's somewhat circular since you have to ask how good is the clock on your "good" system.

You can get the temperature connection by plotting offset and temperature over 24-48 hours.  You can get confirmation by plotting the drift.

Crystals are well known to have a temperature effect so the drift-temp connection shouldn't be surprising.

On Linux, you can get the temperature from
  /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_cur_freq
or similar.  The filename depends slightly on the version of the kernel you are using.

You can get confirmation of the temperature from the disk using hddtemp or smartd and/or by monitoring your room temperature.  (That works better if the disk is in the same box.)

You would really like to measure the temperature at the crystal.  I haven't gone that far.



-- 
These are my opinions.  I hate spam.