[ntp:questions] Vista NTP tamed, but with a blunt object

Dave Hart davehart at gmail.com
Thu Jan 22 05:20:33 UTC 2009


On Jan 21, 8:25 pm, mi... at udel.edu (David Mills) wrote:
> Dave (and others),
>
> On a laptop and where spread-spectrum clock is required, it might be
> better to tolerate a timer-interrut clock, but only if the various
> systematic errors can be averaged out. See the libntp/systeim.c routine
> and the sys_tick variable, which is intended to reduce the residuals for
> thos systems that have only a 100-Hz timer interrupt. This is
> conditioned on the system type define. It should be easy to do the same
> kind of thing for other systems.

Thanks for your response.  ports/winnt/nt_clockstuff.c adj_systime has
much the same logic, though instead of sys_tick, the variable there is
called ppm_per_adjust_unit.  As with the systime.c version, care is
taken to carry over residual from each adjustment.  You can see that
in action with a Windows ntpd.exe built with DEBUG, running with debug
> 2 I believe.  grep (findstr) for SetSystemTimeAdjustment and you'll
note the adjustment units will show a residual pattern like 1 2 1 2 or
6 6 7 6 6 7.

Cheers,
Dave Hart




More information about the questions mailing list