[time-nuts] More 60Hz graphs

Hal Murray hmurray at megapathdsl.net
Fri Feb 7 18:10:11 EST 2014


A histogram of the frequency samples:
  http://www.megapathdsl.net/~hmurray/time-nuts/60Hz/freq-hist.png
The x positions are fudged by 0.02 Hz so multiple dots on the same point 
don't land on top of eachother and hide the ones on the bottom.

Another dip big enough to get below 59.9 Hz:
  http://www.megapathdsl.net/~hmurray/time-nuts/60Hz/2014-Jan-26-a-dip.png

An example of a bug:
  http://www.megapathdsl.net/~hmurray/time-nuts/60Hz/2013-Dec-26-a-bug.png
The kernel code that collects the data and prints it via the /sys file system 
is missing a lock.  Occasionally, the count is off by one.  I assume the 
printout code gets the new time and old count.  That turns one of my samples 
into being low in frequency and the following one being high in frequency.  
The offsets are big enough to stand out.  [I haven't looked at the code to 
verify this.]

This also provides an overall sanity check.  If counts are getting lost due 
to interrupts being locked out for too long, that will stand out like this.  
Similar for extra counts due to noise.



-- 
These are my opinions.  I hate spam.






More information about the time-nuts mailing list