usleep() (in)accuracy

[Boszormenyi Zoltan]

Hi!

I need precise timing and a busy loop on gettimeofday()
is unacceptable because I have to be fair for other
CPU hungry processes. I tried usleep() and nanosleep()
but they are pretty inaccurate. E.g. usleep(7500)
gives 10000-50000 microsecond sleep (19995 on average)
and usleep(20000) gives an average of ~30000 microsecond
sleep. nanosleep() accuracy is the same which is not
surprising because they use the same implementation.

Ingo Molnar's lowlatency patch (2.4.0-C4) did not help.

I tested this on 2.4.0-test6pre1 and a dual 400MHz Celeron
with 192 MB memory.

Can it be solved somehow?

Regards,
Zoltan Boszormenyi <[e-mail]>