cvs commit: src/usr.sbin/dntpd client.c client.h convert.c defs.h ntpreq.c

[Matthew Dillon <dillon@xxxxxxxxxxxxxxxxxxxxxxx>]

dillon      2005/04/24 02:39:27 PDT

DragonFly src repository

  Modified files:
    usr.sbin/dntpd       client.c client.h convert.c defs.h 
                         ntpreq.c 
  Log:
  Change client_check() to calculate the best offset and the best frequency
  correction separately.
  
  The best offset correction now utilizes a standard deviation of the offset
  error with a minimum of 4 samples.  An offset correction occurs when the
  standard deviation is less then 1/4 the average offset.  When an offset
  correction is made, the offset data is cleared (since it becomes invalid
  once a correction is made).  This calculation gives the code an almost
  human perception of when the offset can be safely adjusted, regardless
  of the 'mess' on the internet.
  
  Change tv_delta_micro() into tv_delta_double() since most of the callers
  were (previously) dividing the result by 1E6.
  
  NOTE: the correllation for frequency correction is not working quite as
  well as I had hoped, it's bouncing around a lot for internet ntp sources.
  But the standard deviation on the offset appears to work very well.
  
  Revision  Changes    Path
  1.4       +147 -23   src/usr.sbin/dntpd/client.c
  1.3       +22 -7     src/usr.sbin/dntpd/client.h
  1.3       +3 -3      src/usr.sbin/dntpd/convert.c
  1.3       +1 -1      src/usr.sbin/dntpd/defs.h
  1.3       +1 -1      src/usr.sbin/dntpd/ntpreq.c


http://www.dragonflybsd.org/cvsweb/src/usr.sbin/dntpd/client.c.diff?r1=1.3&r2=1.4&f=u
http://www.dragonflybsd.org/cvsweb/src/usr.sbin/dntpd/client.h.diff?r1=1.2&r2=1.3&f=u
http://www.dragonflybsd.org/cvsweb/src/usr.sbin/dntpd/convert.c.diff?r1=1.2&r2=1.3&f=u
http://www.dragonflybsd.org/cvsweb/src/usr.sbin/dntpd/defs.h.diff?r1=1.2&r2=1.3&f=u
http://www.dragonflybsd.org/cvsweb/src/usr.sbin/dntpd/ntpreq.c.diff?r1=1.2&r2=1.3&f=u