All quartz crystals, such as those used in computer clocks, experience an
effect known as aging
[1][2].
This is a gradual shift in frequency due to outgassing,
propagation of microfractures in the crystal itself, and other factors. Given
an accurate reference frequency to measure against, it should be possible to trace
this effect over time.
Below are graphs of the measured frequency error of three Network Time Protocol
(NTP) servers. For these graphs, the error value has been averaged over each day
to make the data volume managable. Network issues cause spurious high and low
values, but general trends are visible.
NTP Frequency Drift vs. Time
Server: Beltrami
Server: Blizzard
Server: Keyhole
The most data is available for the first server, spanning about 2.5 years, server
two about 1.5 years, and the third server 0.6 years. In all three cases the
drift seems to have changed 0.5 to 2 parts per million (PPM) over the available data.
Compared to the "typical" drift of 5 - 10 PPM per year for an average
crystal oscillator [1],
this appears rather low. It should be noted that all three of these servers
have been running for considerably longer than the span of data available,
so the expected logrithmic dropoff of the aging curve may be responsible.
While long-term trends are arguably indicated by these graphs, other effects on
crystal frequency (such as temperature) make it hard to say definitively that
we are indeed seeing the aging effect. Changes in network topology and traffic loads
can also cause issues with NTP. A 'proper' test would be conducted with a
time interval counter, calibrated against an atomic frequency reference. While
such an experiment would satisfy a true
time-nut, it is beyond
the scope of this webpage.
