[xiph-rtp] My comments

[Tor-Einar Jarnbjo]

Ramón García wrote:

>The idea of Tor-Einar of using a separate data stream for codebooks,
>though smart and interesting, is too complicated. Most software
>frameworks will assume that a single RTP session is all necessary to
>support an audio (or video) stream. Thus this method would make it
>difficult to support Vorbis in DirectShow and other multimedia
>architectures.
>  
>
Are you sure about this? If you combine several content types like 
audio, video and perhaps subtitles, you have separate RTP streams for 
each content type anyway and the RTSP specification explicitly allows 
diffrent streams to be handled by different RTSP sessions.

>Finally, I am against including fragmentation of Vorbis frames. The
>UDP protocol can fragment packets and reassemble, thus there is no
>need to duplicate that functionality.
>
The UDP protocol can't handle packets larger than 64kB, so there must be 
a defined way to split Vorbis packets anyway in the unlikely case that 
the Vorbis packet is too large. Other RFCs for RTP payload formats 
recommend as well, that the payload packets are split into several RTP 
packets if the packet size would exceed the network path MTU. I'm not 
that deep into the IP protocols to determine what's better: rely on the 
UDP implementation to reassemble IP packets or do it yourself.

Tor