[pjsip] Conferfence Bridge Benchmarking

Benny Prijono bennylp at pjsip.org
Wed May 28 08:39:39 EDT 2008

On Wed, May 28, 2008 at 11:19 AM, Thomas Plotkowiak <plotti at gmx.net> wrote:
> Since it is almost impossible to find any data on the internet about the
> scaling of conference bridges I would like to do a test myself with pjsip.

We have confbench.c in samples directory, does it not do what you want?

> When i talk about scaling i mean the CPU usage for mixing audiosignals.
> A chart like Usage of CPU vs. Number of mixed Audiosignals would be perfect.
> Another option could be CPU vs. Number of decoded Audiosignals (or encoded)
> My Problem is: In terms of bandwith a 100Mbit Server would theoretically
> capable of recieving around 6000 16 kbit Streams before he runs out of
> Bandwidth, but i think that it rather runs out of CPU instead with already
> 500 Streams to be mixed in the conference bridge. I cant prove my point
> untill I have some data.

We can do that too of course (by using pjsua, or maybe building
another application to do that), but the problem with that is you're
benchmarking the whole system (codec, jitter buffer, socket, and maybe
even resampling, WSOLA, PLC, and AEC etc.) rather than the bridge
alone, which may or may not be what you want. For benchmarking the
conference alone the confbench.c sample is the perfect start IMO.


More information about the pjsip mailing list