> What is your target arch, what is your time line for experiments? I'm
> expecting to work on the LTTng thing soon again, maybe this week, and
> that would be a good tool for all your periodic-timing-goes-crazy
issues.

I'm running on x86 (geode... puke) and we I can help with experiments at
least daily. I have removed periodic timing (for now) and also set the
scalability stuff to linear, and am very happy with the result. I'm
seeing about 6% of cpu usage in the IRQ0 handler under worst case. I
don't trip the watchdog with the timeout set to 2 seconds like Gilles
suggested.

> 
> One further note regarding overhead: Timers (or timed tasks) running
> over a periodic timebase are expected to have a slightly higher
overhead
> than running (at the same frequency) directly on the master
(==one-shot)
> timebase. That's because the periodic timebase is built _on_top_ of
the
> master base theses days (since Xenomai 2.4). So, unless you can
benefit
> substantially from a jiffies/ticks based timing for your application
> (/wrt internal time calculations or due to the clustering of timer
> events), running directly on the master base is often the more
efficient
> approach.

I don't really need jiffies, but what I do want is there to be a common
timebase for all my tasks in order to implement true RMS scheduling. I
don't want things staggered on creation or after various delays are
inserted into the system.

Thanks for your explanation about the timers, though. I am very happy
with Xenomai's oneshot timer handling. I never could have had the system
playing mp3s with a realtime sound driver and a DAC while controlling a
stepper motor and handling user interaction with oneshot and RTAI (at
least a few years ago.)

Steven


_______________________________________________
Xenomai-help mailing list
[email protected]
https://mail.gna.org/listinfo/xenomai-help

Reply via email to