[ntp:questions] Re: drifting on crystal

Terje Mathisen terje.mathisen at hda.hydro.com
Thu Jan 13 14:51:21 UTC 2005


Marc Brett wrote:
> Instead of OCXOs everywhere, I'd prefer to see a standard connector so consumers
> can easily replace the crystal with a TCXO, OCXO, or one of those new tiny
> atomic clocks if their application requires it.

Yes, please! That's the way it used to be, I can still remember tweaking 
1984-vintage IBM ATs by replacing the 6 MHz MB crystal with either a 
fixed 8/9 MHz one, or by a tunable oscillator. :-)

TCXO is an interesting tradeoff, particulary since people have tried to 
make their own:

Take a standard motherboard with one or multiple temperature sensors 
(typically used to monitor cpu/disk/graphics units), and place the 
sensor on the crystal.

Run with a good local reflock (i.e. GPS) for several days, and determine 
a best-fit approximation for the relationship between delta temperature 
and delta drift rate.

Using such a curve, precompensate the local clock by the actual current 
temperature, i.e. use this each second to tweak the clock adjustment 
calculated by the rest of ntpd.

I remember seeing one such setup that resulted in an order of magnitude 
less drift/jitter for the server.

Terje
-- 
- <Terje.Mathisen at hda.hydro.com>
"almost all programming can be viewed as an exercise in caching"



More information about the questions mailing list