Re: [rtcweb] DRAFT Agenda for RTCWEB

[Harald Alvestrand <[e-mail]>]

On 03/01/2013 07:57 PM, Dan Wing wrote:
>> -----Original Message-----
>> From: [e-mail] [mailto:[e-mail]] On Behalf
>> Of Dale R. Worley
>> Sent: Friday, March 01, 2013 7:59 AM
>> To: Ejzak, Richard P (Richard)
>> Cc: [e-mail]
>> Subject: Re: [rtcweb] DRAFT Agenda for RTCWEB
>>
>>> From: "Ejzak, Richard P (Richard)" <[e-mail]>
>>>
>>> Since multiplexing of the data channel with RTP media has been shown
>>> as a desirable feature of BUNDLE (and most of its variants), I would
>>> suggest that this be treated as a significant advantage for BUNDLE
>>> (and similarly capable variants) over any proposal without it.
>>> Cullen's "Plan A" is preferred over Plan B precisely because it has an
>>> incremental muxing advantage.
>> As far as I can tell from my analysis
>> (http://tools.ietf.org/html/draft-worley-sdp-bundle-04#section-8)
>> SCTP-over-DTLS can be demuxed from RTP and STUN quite easily.  (This
>> comes from RFC 5764 section 5.1.2.)  And SCTP can be demuxed from the
>> rest as long as you control the range of SCTP ports used.  (And since
>> the ports aren't actually to route the packets to the receiver (the
>> underlying UDP does that), you have freedom in choosing SCTP ports.)
>>
>> So I don't see anything blocking Plan A as compared to Plan B.  Of
>> course, we have to *do* a bundle technique, but we've got a large
>> library of possibilities now and can look at the fundamental design
>> questions in context.
> The analysis should also examine impact of port-based QoS on
> multiplexing data over the same port as audio.  Even audio and
> video often prefer different drop precedence.  I know the general
> consensus is to just supply more bandwidth, yet we consume all
> available bandwidth much like disk space, memory, and CPU/GPU
> cycles.
>
A reason for the particular way BUNDLE is designed is so that it permits 
you to bundle all the audio separately from all the video if that's your 
desire.
It just doesn't force you to do so.

Once we have BUNDLE in place and the m-line thing settled, we can return 
to the issue of what the default operational mode should be.