Re: [libmicrohttpd] perf_get_concurrent results seems weird

[Christian Grothoff]

hi!

You need to understand that the benchmark is by default set to run a
rather small number of iterations (500 rounds), so on machines that are
not super-slow, you will get highly inaccurate results. To really
benchmark something on your system and get meaningful results, you may
need to adjust the CPU_COUNT and ROUNDS #defines in the source code.

Happy hacking!

Christian

On 08/27/2015 10:41 AM, Allan Wind wrote:
> Hi,
> 
> I was just running perf_get_concurrent and I noticed that in this 
> run 3 of the tests have identical results, and the exact 25k 
> seems weird too:
> 
> Parallel GETs using internal select: 16129.032258 requests/s
> Parallel GETs using thread with select: 23255.813953 requests/s
> Parallel GETs using thread pool with select: 23809.523810 requests/s
> Parallel GETs using external select: 25000.000000 requests/s
> Parallel GETs using internal poll: 25641.025641 requests/s
> Parallel GETs using thread with poll: 23255.813953 requests/s
> Parallel GETs using thread pool with poll: 24390.243902 requests/s
> Parallel GETs using internal poll: 27777.777778 requests/s
> Parallel GETs using thread pool with poll: 23255.813953 requests/s
> 
> ... and this 2nd run, identical numbers from the previous run but from a
> different test:
> 
> Parallel GETs using internal select: 16666.666667 requests/s
> Parallel GETs using thread with select: 23809.523810 requests/s
> Parallel GETs using thread pool with select: 22222.222222 requests/s
> Parallel GETs using external select: 25000.000000 requests/s
> Parallel GETs using internal poll: 26315.789474 requests/s
> Parallel GETs using thread with poll: 23809.523810 requests/s
> Parallel GETs using thread pool with poll: 25641.025641 requests/s
> Parallel GETs using internal poll: 27027.027027 requests/s
> Parallel GETs using thread pool with poll: 24390.243902 requests/s
> 
> 
> /Allan
>

signature.asc
Description: OpenPGP digital signature