CPU load in voice mode
The results of this test can help you to choose CPU for your Asterisk system. For this test we used two Debian boxes with 2.6.15 smp kernels and Asterisk 188.8.131.52 installed. On PC 1 we installed zaptel-1.2.6 and quasar driver version 1.9.1. Quasar board driver was slightly modified to provide opportunity to emulate off-hook events on arbitrary channel. Zaptel channels in zaptel.conf were configured as fxols. On sip channels we used alaw in all experiments to reject transcoding overhead.
Experiments: Using small script on PC 1 we emulate off-hook events on some Quasar's zaptel channels, according to the dialplans Asterisk on PC 1 start to establish sip connections with PC 2 and Asterisk on PC 2 begin to execute the Playback application on corresponding channels. As a result we were able to listen sound generated by Asterisk on PC 2 using external FXS module. Every measurement we control sound quality using this module and phone on some random spans and timeslots. To measure CPU load we used vmstat (vmstat 1 -n > file_to_put_results) utility. Statistics were collected for 2-3 minutes for every measurement. We've made two series of measurements with echo canceler (KB1) enabled and disabled.
On the following graph you can see the results of the measurements:
This graph is for 2,41 GHz P4 processor with HT enabled. Using this graph you can estimate for example that to build simple sip gateway which can handle 240 simultaneous voice channels (8 E1 ports of Quasar-8PCI) without echo cancellation and transcoding you need 3 GHz P4 HT or better CPU (assuming reserve of productivity).