the uplink to my nearest upstream servers is quite good and thus temperature
changes (especially due to the air-conditioning now) become the main source
of an offset scatter. An interesting approach described on
http://www.ijs.si/time/temp-compensation/ requires a change in the ntpd's
source code. The page http://lx.ujf.cas.cz/ntp-lx/temp describes an
experiment based on the ntptime utility and differential frequency
corrections. I should say that although there is some improvement, my doubts
about still remain.
Karel Sandler