Matthew Mondor wrote:

Ignoring everything else...

> $ time ./clocktest1 1000000 
>     8.35s real     0.52s user     7.81s system
 [ gettimeofday ]
> $ time ./clocktest2 1000000 
>     8.50s real     0.52s user     7.96s system
 [ clock_gettime ]
> $ time ./clocktest3 1000000 
>     3.40s real     0.29s user     3.09s system
 [ kern.hardclock_ticks sysctl ]

Note that the kern.hardclock_ticks sysctl simply reads a kernel variable
and returns it without doing any extra work.  That variable is updated
once per clock tick (obviously!).  This will have a HZ granularity
(0.01 second for i386 by default?).

gettimeofday() (and I assume clock_gettime() from how long it takes)
will call microtime() or some variation thereof and actually give you
as high as resolution as possible of the current time, which will be
sub-microsecond on modern i386.

So no surprise that the sysctl program is faster...

Cheers,
Simon.