FreeSWITCH Capacity Testing Needed More Than a Calls-Per-Second Number

Calls per second is an attractive benchmark because it produces one number. It can also hide the resource that actually limits the system.

A signalling-only call that connects and clears quickly stresses transaction processing differently from a ten-minute call with RTP, recording or transcoding. Concurrent-call capacity depends on memory, sockets and media work. CPS depends heavily on setup-path CPU and database or application latency.

I started defining scenarios before measuring: registration load, short signalling calls, long pass-through media, codec conversion, and failure bursts. For each scenario I watched CPU, scheduler delay, memory, file descriptors and SIP response distribution.

The useful capacity number is therefore tied to a workload. 'This server handles X CPS' means little unless I know what each call required and what failure threshold defined the end of the test.