[ntp:questions] Idea to improve ntpd accuracy

Hal Murray hmurray at megapathdsl.net
Sun Feb 28 04:18:38 UTC 2016


> I thought of that too but rejected it as way too machine-specific.  Because
> the clock PLL does not track frequency ramps feed-forward could  be very
> effective (e.g. order of magnitude). To make this work however,  you need to
> first discover which one of many potential temperature  sensors on the MB
> are most closely correlated with frequency, then  measure a rough linear
> gain from temperature to PPM. Seemed too  hardware-instance-specific to me,
> but perhaps there's a way... 

You want to measure the temperature of the crystal.

Several/many years ago, Linux used to use the RTC/TOY clock for the main 
timekeeping.  Now, the main CPU clock is used.  You want to get a temperate 
probe on that.


Mark Martinec has a wonderful web page on that.
NTP temperature compensation
  https://www.ijs.si/time/temp-compensation/


-- 
These are my opinions.  I hate spam.





More information about the questions mailing list