Re: [Xenomai-help] Reducing Kernel Mode Interrupt Latency

[Philippe Gerum <rpm@xenomai.org>]

Henry Bausley wrote:
> Is there any way the kernel mode interrupt handling maximum times can be
> decreased.  I very much need to reduce my kernel mode interrupt
> latencies.  I know my hardware is capable of better performance based
> upon measurements done with RT Linux.  Below is our timing comparison,
> all times are the maximums.
> 
> 
> Xenomai 2.4.5 interrupt handler  w/ Kernel 2.6.26 powerpc, gcc 4.2
> -------------------------------------------------------------------
> AMCC 440EP - 24.5 usec
> AMCC 460EX - 16.5 usec
>

While running long performances tests on a Sequoia board here (I'm still
investigating the latency trace data and evaluating optimizations), I've never
been able to reproduce such worst case latency values.

The figures I have on the 440EPx (w/ 2.6.26 + Xenomai 2.4.5 + I-pipe 2.2-07 +
ELDK-4.2 toolchain) are:

klatency: 15.2 us
latency -t2 : 17.1 us

latency -t1: 26.1 us
latency -t0: 34.9 us

/proc/xenomai/latency reset to 0 (no timer shot anticipation)
continuous process creation loop and interrupt flood on entry.

The figure you are most likely interested in - the one that actually compares
with RTLinux's latency test - is given by the klatency test, i.e. pure in-kernel
timer handler, no user-space Xenomai thread). Could you tell us a bit more about
your kernel config, and the kind of stress you put on the system? TIA,

> 
> FSM Labs RT-Linux 2.2 w/ Kernel 2.6.14 ppc, gcc 3.3
> ---------------------------------------------------
> AMCC 440EP - 10.0 usec
> AMCC 460EX - Not Available, would require 10's of thousands of dollars
> to WindRiver
> 
> 
> 
> 
> _______________________________________________
> Xenomai-help mailing list
> Xenomai-help@domain.hid
> https://mail.gna.org/listinfo/xenomai-help
> 


-- 
Philippe.