On Aug 12, 2013, at 3:40 AM, pablo platt <pablo.platt@gmail.com> wrote:

Hi,

I'm currently using a RTMP server for audio conference. There are no performance or latency issues because there is no encryption, transcoding or mixing involved. A 1GB VPS server can handle more than 100 users without a problem.

I'm trying to evaluate FreeSWITCH so I'll be able to support both Flash and WebRTC users in a conference.

My use case is two types of conferences: - A conference with up to 5 participants all speaking. - A conference with 20-30 participants with 2-3 speakers.

What part require more CPU? - Encryption (DTLS-SRTP) - Transcoding - Mixing

I'm pretty sure it will be a toss up between srtp (the dtls-srtp negotiation will be minimal in comparison) and transcoding depending on the codec. g711 will be for sure less than srtp. g729 will be more than srtp. The mixing is typically pretty minor on cpu in comparison depending on the number of talking participants, which is typically 0 or 1.

Does FS mix a channel separately for each participant or can it reuse the same mix?

It does one mix, but for transcoding it does a separate encode for each participant.

For example, if I have 1 speaker and 20 listeners, 10 with speex and 10 with opus, does FS produce 2 streams or 20?

Does FS exclude silent participants from the mix to save CPU?

Yes. This does require us to decode all incoming streams not marked as muted.