Correction: A good frequency counter told me that even on kernel 2.4.4 my soft rt 60Hz framerate was averaging 17.000msec per period. What was misleading was that on 2.4.4 the invocation of the alarm handler jumps around, and occasionally hits its precise period (a beat frequency w/ timeslicing?).
Whereas on 2.2.17, an oscope shows the same handler happening on even multiples of my 1msec timeslices - same aggregate frequency, less jitter. -S steve rosenbluth wrote: > > Apparently kernel 2.4 gave priority to software interrupts (signals) so > that itimer() periods are more accurate...(someone please explain) > > The net result is that soft real time periodic jobs are now more > accurate. > Using this: > > kernel compiled for 1000 timeslices/sec > signal(SIGALRM, handler_func) > setitimer(ITIMER_REAL, &interval, 0) > sched_setscheduler(THIS_PROC, SCHED_FIFO, &my_sched) > > We get periods of _fractions_ of milliseconds on k2.4.4 , > Whereas on k2.2.17, periods were in multiples of 1 millisecond. > > I have an application which uses the above to request alarm handlers at > 60 Hz (a 16.66 msec period). On k2.2.17-rtl3.0pre9 the period of the > handler (seen on an oscilloscope) was 17 msec. On k2.4.4-rtl3.1pre3 the > period of the handler is 16.6 msec. > > (I'm assuming its the kernel doing this, and not the standard rtlinux > patch.) > > -- Steve Rosenbluth > > -- [rtl] --- > To unsubscribe: > echo "unsubscribe rtl" | mail [EMAIL PROTECTED] OR > echo "unsubscribe rtl <Your_email>" | mail [EMAIL PROTECTED] > -- > For more information on Real-Time Linux see: > http://www.rtlinux.org/ -- Steve Rosenbluth Jim Henson's Creature Shop - LA 2821 Burton Ave Burbank CA 91505 818-953-3030 -- [rtl] --- To unsubscribe: echo "unsubscribe rtl" | mail [EMAIL PROTECTED] OR echo "unsubscribe rtl <Your_email>" | mail [EMAIL PROTECTED] -- For more information on Real-Time Linux see: http://www.rtlinux.org/
