[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
- Date: Tue, 13 Jul 1999 07:54:47 -0500 (CDT)
- From: joel at OARcorp.com (joel at OARcorp.com)
- Subject: RTEMS performance
On Tue, 13 Jul 1999, Ramon Costa [iso-8859-1] Castell? wrote:
> When I get information about a RTOS some characteristic times
> are usually given to me. These times represent the performance of the
> (Interrupt Latency, Context Switch Times, ....)
> These times are usually machine dependent. My question is these times
> are obtained through an experimental set-up or are based in the RTOS
> knowledge and the instruction execution times of the
> target machine.
> Which are the times for RTEMS 4.0 on a Pentium-II PC ?
> If they are based on an experimental set-up I would like to get
> or pointers to the description of the methodology, Is the methodology
> standard or it is adhoc for each system ?
It is standard. There is a set of CPU supplements which describe this.
Basically, the set of tmtests included in the release time everything in
the Classic API (rtems/src).
The test tm27 times interrupt latency except for maximum interrupt disable
time which we have always calculated by hand.
Lately it has gotten harder to hand count cycles and we just have counted
instructions when it matters enough to do it. The worst interrupt disable
time is normally in one of two places based on the relative cost of a bit
scan or memory instructions. For example, on the i960 bitscan is one
cycle so memory ops out weigh it and the worst interrupt disable period is
in _Thread_queue_Enqueue_priority. On the m68k and i386, bitscan is
expensive so it is in the latter part of _Thread_Set_state where
_Thread_Calculate_heir does 2 bitscan operations.
If you want the times for a particular machine, the tests are there. Just
run them. :)
Running the tmtests is a nice way to see if the performance of your target
board is in the range it should be. If you see times 10x off of where
they should be, then something is wrong -- too many wait states, clock is
wrong, caching is disabled, etc, etc.
Joel Sherrill, Ph.D. Director of Research & Development
joel at OARcorp.com On-Line Applications Research
Ask me about RTEMS: a free RTOS Huntsville AL 35805
Support Available (256) 722-9985