RTEMS mvme5500 bsp
Kate Feng
feng1 at bnl.gov
Wed Jul 27 20:31:13 UTC 2005
Robert Varner wrote:
>
> Thank you for your help last week in understanding interrupts in RTEMS
>
> on the MVME5500. With your suggestions and a little conceptual break
> through we have VME interrupts working well.
>
> Have you or anyone you know of made benchmarks of the performance of
> RTEMS on the MVME5500? Particularly, we would like to know enough to
> judge which services to use and the time required for various context
> switches.
>
> Robert Varner
> ORNL Physics Division
I finally got around to optimize the IRQ handling and
prioritization. Many thanks to Till Straumann of SLAC
for providing some suggestions. I modified (for MVME5500)
the latency benchmark software which Till posted
on http://www.slac.stanford.edu/~strauman/rtoslat/
The test was performed on the MVME5500 board. 2,000,000
timer interrupts were generated at a rate of 4KHZ. The
maximum and average letencies for both the idle and loaded
system were measured. The loaded system was subject to heavy
network and console activities. Please note that the worst case
(e.g. maximum) is critical for a real-time system.
The result is listed below. Based on the new algorithm, the result
for a VME interrupt in RTEMS should be very close to what is listed
as long as it's set at the highest priority in irq_init.c. Please let
me
know otherwise. Any suggestions or comments before I submitted
the patch ? Also, the updated software fixed the bug of
the irq->on() API, which was reported back in Feb.. -- Kate
______________________________________________________
MVME5500 Interrupt Latency | Context Switching
_________________________________________________________
max (average) | max (average)
_________________________________________________________
Idle System: |
RTEMS 5.04 (3.45) | 6.80 (0.96)
vxWorks 12.13 (1.58) | 18.68 (0.91)
_________________________________________________________
Loaded System:
RTEMS 8.17 (3.74) | 24.88 (1.69)
vxWorks 22.49 (2.04) | 40.76 (4.1)
_________________________________________________________
** All units are in usec.
More information about the users
mailing list