Rhealstone detailed benchmark

Andreas Kölbl andreas.koelbl at st.oth-regensburg.de
Mon May 29 16:35:04 UTC 2017

Am 28.05.2017 um 18:24 schrieb Joel Sherrill:
> On Sun, May 28, 2017 at 6:43 AM, Andreas Kölbl <
> andreas.koelbl at st.oth-regensburg.de> wrote:
>> Hi guys,
>> I've tried measuring the real time capability of a BSP using the
>> Rhealstone benchmark. This benchmark only tells about the average time
>> spent for e.g. task switching.
>> Is there interest in a more fine-grained Rhealstone benchmark, which
>> prints the Worst Case Execution Time (WCET) or the deviation of the
>> measured values?
>> Has there already been an attempt to modify Rhealstone in order to get
>> the WCET from each benchmark?
> Rhealstone was a proposal from someone at Intel in 1989 that appeared
> in Dr Dobbs Journal.
> http://www.drdobbs.com/article/print?articleId=184408081&siteSectionName=
> There was a follow up article a year later with some code which I didn't
> remember. I am not sure how closely the code written for RTEMS follows
> that code.
> http://www.drdobbs.com/article/print?articleId=184408332&siteSectionName=cpp
> AFAIK there is no "standard" source for this benchmark -- just the
> definitions
> and some odd code for a few RTOSes.
> We would welcome a review of the code against the original articles
> and the code they published as well as improvements in the numbers
> gathered and their reporting. Statistical information and WCET would
> be valuable if the code accurately reflects the intended measurement.
I will try this for each benchmark and respond if I were successful.
The source of the benchmarks (drdobbs) seems to be similar to the ones
used in RTEMS.
So maybe there is a reason, like to get a more efficient implementation
(with caches) or more comparable results when measuring the hardware
this way?
> One thing I have noticed in benchmarks over the years is that they
> may want to measure say blocking time of a semaphore with a
> set of N threads. But they accidentally include the code associated
> with the first execution of each thread. So you have to be careful
> to not include unintended overhead.
> Sorry for the long answer but if you are interested in these benchmarks,
> you are encouraged to help improve them.
Sure ;)

Andreas Kölbl
> --joel
>> --
>> Andreas Kölbl
>> _______________________________________________
>> users mailing list
>> users at rtems.org
>> http://lists.rtems.org/mailman/listinfo/users

More information about the users mailing list