[EMAIL PROTECTED] wrote:
> In the meantime I modified the benchmark with a call to the routine
> pthread_set_mode_np(0, PTHREAD_SHIELD) immediately after thread is
> created, and observes a huge reduction in execution jitter, but at
> the expense of wakeup jitter. So I concluded the interrupt shield was
> definitely not doing much without this call. Am I wrong ?
I guess Philippe and Stephane gave you answers regarding the way to
activate the shield (I did not check if pthread_set_mode fails when the
shield was disabled at compilation time and is used at run time, if not,
maybe it would be a good idea ?).
Now about the wakeup jitter, if you are still using the "cruncher" as in
RTAI/Fusion, what is called "nanosleep jitter" may not be a precise
measurement of the wakeup latency. A better measurement of the
scheduling latency using the POSIX skin is obtained using the
clock_nanosleep call with the TIMER_ABSTIME flag, pass an absolute time,
and measure the difference between this time and the wakeup time.
Also note that CLOCK_MONOTONIC will give better accuracy than
CLOCK_REALTIME, since it is using the tsc even when system timer is in
periodic mode.
--
Gilles Chanteperdrix.
