[OpenSIPS-Devel] [OpenSIPS-Users] Code profiling for OpenSIPS 1.6

[Bogdan-Andrei Iancu]

Hi Stan,

Stanisław Pitucha wrote:
> 2009/10/8 Iulia Bublea <iulia at opensips.org>:
>   
>> I uploaded on the opensips.org web site (see
>> http://www.opensips.org/Resources/TestsProfiling) some results and
>> interpretations of a code profiling for OpenSIPS 1.6.
>>     
>
> I think there's a small mistake on the page: "... you can see that
> managing the SIP messages (parsing and building) takes more than 50%
> of the CPU usage for OpenSIPS." but the graph shows 34% (parsing) +
> 11% (building) == 45%.
>   
Right , fixed the sentence :)
> Just curious - did you also generate any statistics
> per-config-function (just to see how long does each operation take)?
> They would have to ignore the parsing time probably, as it will not
> happen on every call and will be config-dependent. (I know I can get
> that information from the gprof report myself, but I hope you've done
> it already ;) )
>   
some function do no even show up in the list (as maybe they have a very 
very low CPU usage and they were never hit by profiling sampler).

of course you can generate endless reports based on the profiling values 
- this is why we decided to publish the gprof output also, so anybody 
can use it for his own statistics. If you generate some other reports, 
we can add them on the web page, no problem.
> Also - it seems to me like 16% for mem ops is a pretty high value. Do
> you think it's a reasonable part? (same as above, I'll actually check
> that myself, but wanted to know your opinion)
>   
well, 3 notes here:
1) memory debugger was enabled during the profiling, so there is an 
extra burn
2) the mem debugger uses a differeng mem manager (QM) which is slower 
than the default one (F_MALLOC)
3) the result shows both pkg and shm, as there is the single mem manager 
(set of functions), but different pools of mem,

Regards,
Bogdan