[OpenSIPS-Devel] [OpenSIPS-Users] OpenSER - RTPproxy Benchmark Test

[Maxim Sobolev]

Alex Balashov wrote:
> I'd be curious to know if you get better results using MediaProxy 2.0, 
> now that it uses kernel packet routing hooks for the RTP switching.
> 
> 750 is quite disappointing.

Alex,

I think before calling it "disappointing" you should note the following 
few things:

1. The tests were performed using the old 1.0 version. In the latest 
stable version 1.1 we have made several important performance enhancements.

2. This was a synthetic test, therefore its results should be taken with 
the grain of salt, as it might not accurately simulate real world 
conditions.

3. In order to take advantage of dual-core architecture, some extra 
effort is necessary (see http://www.rtpproxy.org/wiki/RTPproxy/FAQ).

Just to give you indication of the real-world numbers I just checked one 
of the production systems our customer use. The 4-core 3GHz Xeon system 
routinely pushes around 40-60MBps of RTP traffic and runs 4 copies of 
RTPproxy in parallel. Each of 4 copies has since the last boot forwarded 
about 30,000,000,000 RTP packets and used some 150,000 seconds of CPU 
time. As you can easily calculate, the peak packet forwarding capacity 
should be 4 x 200,000 packets per second, that is 800,000 RTP packets 
per second per whole box. With the shortest possible 10ms packets, this 
should allow for max 8,000 sessions, probably even more with 20 or 30 ms 
packets. Even if you discount that peak number heavily by 20-30% (to 
account for possible CPU time accounting errors in the Linux kernel and 
other scalability issues), still you will get conservative estimate of 
4,000-5,000 sessions per server.

Regards,
-- 
Maksym Sobolyev
Sippy Software, Inc.
Internet Telephony (VoIP) Experts
T/F: +1-646-651-1110
Web: http://www.sippysoft.com