[xiph-rtp] Updated Vorbis and new Theora I-D's

Ralph Giles giles at xiph.org
Fri Feb 4 10:53:25 PST 2005


On other comment after talking to Jack last weekend.

In section 4.1 you describe in-line transmission and at least imply
that periodic in-band retransmission is allowed. I thought this section
should be expanded slightly to make the following points:

  Encoders/Servers MAY send the headers in-band at the start of the
  stream and MAY periodically resend them or send new headers when
  the decode parameters change. Clients MUST handle these inline
  headers if they are received. Server developers should be aware
  however than the bandwidth consumed by the headers is disruptive
  in many applications and are encouraged to rely instead on out-
  of-band transmission of the headers.

Jack was in favor of no in-band transmission at all, so the idea
is to discourage this in the general case. But I wanted really
dumb servers to be able to just send the packets as they come
in reliable network conditions. Periodic retransmission builds
on top of that.

As written, it sounds like you can just send any header at any
time, although getting info and setup headers separately would
be a pain if we're hashing them together. (I guess you'd have
to build a cache entry for every combination whether it made
sense or not.) Is it worth clarifying that while metadata packets
can occur on their own, info+metadata+setup headers must be
sent all three together contiguously?

 -r


More information about the xiph-rtp mailing list