[ntp:questions] Re: Fingerprinting hosts by clock skew
David L. Mills
mills at udel.edu
Fri Mar 11 02:21:45 UTC 2005
We did build a special precision clock for the Sun Sbus several years
ago. More recently Poul-Henning Kamp built one based on a Rubidium
oscillator. Our design was based on a counter/devider that could be
disciplined by changing the divide ratio at intervals determined by
another counter. I expect P-H's design does the same thing. In
principle, a similar design could be incorporated on the motherboard,
but there is little incentive to do that.
There are other ways to control the frequency, including a variable
frequency crystal oscillator (VCXO) controlled by an external DC signal
and DAC. An innocent motherboard might be hacked to do this, but with
modern surbace-mount components that might be harder than it looks.
The nanokernel modifications in FreeBSD come pretty close to VCXO
performance, if only the clock oscillator was more stable. Some radios I
have have optional VCXO or TCXO drop-in replacemnts. The only cost to
the manufacturer is to have a plug-in crystal that can be swapped for a
good hot rock.
> Brian Inglis writes:
>>Would be easier to attach the case thermal sensor to the clock
>>generator, wherever that is nowadays.
> Why not just build hardware RTCs that allow for extremely fine
> adjustments via software? NTP could calculate the correct adjustment,
> then program the RTC hardware directly, ultimately producing an
> extraordinarily accurate hardware clock. A clock synchronized in this
> way would also eliminate fingerprinting by clock skew, since the skew
> would soon fall to zero.
More information about the questions