Re: [rtcweb] #15: Section 4.8: SSRC signaling

You can easily exhaust the dynamic payload type number space (96-127) with only a few media lines, not 30. Keep in mind the primary purpose of payload type is to negotiate codecs, not demux different streams. An offer from a highly interoperable product will contain many payload types for each stream. It is not uncommon to support 8+ payload types and 4+ streams, which quickly exhausts the dynamic space. Also keep in mind that many codecs have different configurations that are expressed as different payload types, so even supporting a single audio and video codec can yield many payload types if those codecs are highly configurable.

Consider a very simple product (smartphone) in a very simple use case (3-way video chat). It supports several audio codecs (OPUS, AAC-LD, AMR-WB, AMR-NB, G.711) and several video codecs (VP8, VP9, H.264, H.265). It supports several H.264 profiles (BP, HP, SBP, SHP) and packetization modes (0, 1). This already exhausts all 32 dynamic payload types, before we even consider things like FEC, scalable or simulcast layers sent via MST, screen sharing, multiple cameras, etc.

So I would strongly discourage folks from assuming payload type demux is viable in most use cases. It may be viable when there is a single audio and video stream, each using a single codec and configuration. But it quickly becomes impractical or impossible when the product gets more capable or the use case gets more interesting.

Mo

-----Original Message-----
From: rtcweb-bounces@ietf.org [mailto:rtcweb-bounces@ietf.org] On Behalf Of Dale R. Worley
Sent: Tuesday, April 23, 2013 2:58 PM
To: rtcweb@ietf.org
Subject: Re: [rtcweb] #15: Section 4.8: SSRC signaling

> From: Martin Thomson <martin.thomson@gmail.com>
> 
> On 22 April 2013 15:15, Dale R. Worley <worley@ariadne.com> wrote:
> > My understanding is that associating an incoming RTP packet with an m=
> > line is a solved problem -- the transport association on which the
> > packet arrives determines the bundle, and (within all of the currently
> > active bundling proposals) the payload type tells which m= line within
> > the bundle.
> 
> Not so.  If you have five PTs on each m= line, and 30 m= lines, you
> just don't have that many PTs available.  At some point you have to
> recognize that SSRC is what you have to use.  (Neither number is
> extraordinary.)

But why would you use 30 m= lines?  You'd only have 30 streams in a
videoconference-type situation, where the streams don't have distinct
roles and so don't need to be given role labels.  As far as I know,
current videoconference systems don't put each video stream in a
separate m= line.

Dale
_______________________________________________
rtcweb mailing list
rtcweb@ietf.org
https://www.ietf.org/mailman/listinfo/rtcweb