Hello, I believe the timers are disabled by setting <param name="rtp-timer-name" value="none"/> in the sofia profile(s). The default configurations use a value of "soft". I have similar performance concerns bridging G711 SIP calls between two sofia profiles on an AMD Geode processor. With full media it occupies roughly 3.0% CPU per call. It scales to 25 calls with room for other basic system processes. Maybe 26 or 27 if we really pushed it and maintained the cps extremely low. Proxy media mode gives about a 15% performance increase in our case. Bypass mode was not tested because it isn't appropriate for our application. - Jeff On Fri, Feb 8, 2013 at 7:21 AM, Varghese <p.varghese@cem-solutions.net>wrote:
Hi all,
Any update on this issue. Any other options we need to try with freeswitch.
E1 PRI with freetdm freeswitch in freescale and cpu load is going to 100 with 15 calls.
Thanks And Regards
Varghese Paul
Hi Anthony,
Thanks for the information.
Kernel is Linux-2.6.32 patched by Freescale team for the hardware.
Further we tested the freeswitch-1.2.6 on Freescale and following are observations.
We set the ulimit options as per the wiki link and ulimit –s 240
1. With switches *-np* and *–lp* the result is same i.e. maximum of 90 calls and top command shows 100% CPU for freeswitch process
2. With RTP Bypass mode we could make 250 calls and CPU usage is 20%
3. With Proxy media mode, and default rtp mode the CPU is very high. Approximately 1 to 1.5% per call and scales up to 90 calls
4. Set the option enable-timer value = false in SIP profile, stun-enabled=false given all IP address in ext-rtp, rtp-timer-name=none/soft etc., still the results are same
Q1) Can you explain “asterisk also uses blocking reads in its rtp stack where we have timers. You can disable the timers as documented in the wiki” which parameter you are referring to and where we need to disable timers.
Q2) Do we need to try any other options to increase the performance numbers?
F.Y.I, We used Empirix Hammer SIP Call Generator for testing SIP calls.
Thanks Varghese Paul
Anthony Minessale wrote:
That depends on the version of linux, the kernel version and several other factors.
Were you actually moving media when testing, what tool are you testing with?
On 32 bit you should make sure you have the stack size at 240 (ulimit -s 240 before starting) Also you may need to add the -lp or -np startup flags to reduce the realtime threads.
asterisk also uses blocking reads in its rtp stack where we have timers. You can disable the timers as documented in the wiki..
On Tue, Feb 5, 2013 at 6:37 AM, Varghese <p.varghese@cem-solutions.net <mailto:p.varghese@cem-solutions.net>> wrote:
Hi all, We are facing following observations/issues with freeswitch on freescale processor boards. We ported freeswitch with freescale provided tool chain. _*Freeswitch Configuration: version 1.2.6 (git version)*_ 1. Only enabled mod_sofia, mod_dialplan, mod_command,mod_dptool and mod_console. 2. set the ulimit options and followed the steps in the following link
http://wiki.freeswitch.org/wiki/Performance_testing_and_configurations 3. SIP test tool run the load test for 500 users in freeswitch and configured the dial plan to just bridge the channels 4. Freeswitch media processing is in default mode: i.e media will go through freeswitch _*Freescale Board Configuration:*_ CPU : 1GHZ PowerQUICC III, 32 bit RAM: 1 GB Linux- 2.6.32 Load test is conducted for SIP to SIP calls with RTP. _*Observations:*_ Linux TOP command shows increasing CPU usage per call and increases to 100% with only 90 calls. After that system response is very slow We could also found delay in media.
_Surprisingly, with Asterisk 1.8 the same system configuration works 250 calls without any problems. _ _*Questions: *_1. Any one ported Freeswitch on Freescale or any embedded processors? Any performance figures or references will be appreciated 2. How many simultaneous calls can be possible with 1GHz PowerQUICC III processor ? 3. Any more settings are required in freeswitch or linux for scalability?
Thanks And Regards
Varghese Paul