[ntp:questions] Re: Fingerprinting hosts by clock skew

Michael Deutschmann michael at talamasca.ocis.net
Wed Mar 9 04:07:02 UTC 2005


On Sat, 5 Mar 2005, David L. Mills wrote:
> However, the most useful property of the clock frequency is to track
> systematic changes, which serves as a quite accurate thermometer.
> [...] we are talking about millidegrees here.
> 
> [...] I should add to that proposal using NTP frequency to
> reveal outdoor temperature if the IT staff would put the computer
> outside the building in a doghouse on the roof.

Due to the difficulties of calibrating to real world units, I don't think
this would be practical.  A sensor such as a thermistor can be waterproofed
and taken through it's range in a water bath.  You can't do that to a
computer, especially if it's on.

A better use for the phenomena would be a simple hardware mod to make NTP
systems more stable.  Find the oscillator can on the motherboard, glue a
50kOhm-nominal thermistor to it, and then run leads from it out of the case
and then back to pins 1 & 3 of the joystick port.  Then (the tricky bit...)
alter ntpd to search for a correlation between the "Joystick 1 X axis"
reading and the frequency correction, and start actively compensating when/if
the estimated curve firms up sufficently.

If you really can take millidegree accurate readings of temperature via the
drift, that would imply that nearly all the noise in the drift is from heat,
and the above approach could make a large improvement.  Of course someone
would have to work out how to integrate it sanely into the rest of the NTP
algorithms...

---- Michael Deutschmann <michael at talamasca.ocis.net>



More information about the questions mailing list