Re: [MMUSIC] SCTP question: Where does it multiplex?

[Paul Kyzivat <[e-mail]>]

Harald - comment at end.

On 12/6/12 8:02 AM, Harald Alvestrand wrote:
> On 12/03/2012 09:13 PM, Paul Kyzivat wrote:
>> On 12/3/12 6:13 AM, Harald Alvestrand wrote:
>>> On 12/03/2012 10:02 AM, [e-mail] wrote:
>>>> Hi,
>>>>
>>>> I agree that it would be useful to be able to run data channel on the
>>>> same 5-tuple as RTP/RTCP, and even making it the default and
>>>> must-implement way for the browsers.
>>>>
>>>> QoS/prioritization comes to mind as the only potential drawback. The
>>>> endpoint can naturally still give different priorities to the
>>>> different components (audio, video, "data") it is sending. I assume
>>>> SCTP also has TCP-style receiver window that can be used to throttle
>>>> reception if needed. And presumably RMCAT will find solutions to
>>>> congestion control in genenral. But from the network perspective there
>>>> is only one flow, and the only(?) way to differentiate the components
>>>> is to use DSCP. I don't think our QoS (or API) draft yet clearly
>>>> explains how DSCP's are learned and set, or whether that is even going
>>>> to be a realistic solution.
>>>
>>> That is indeed the only reason I can think of to specify multiple
>>> m-lines for DTLS/SCTP (and we don't need to bundle them together).
>>>
>>> At the moment, the MMUSIC API doesn't define a way to control which
>>> m-line a DTLS assocation gets mapped onto.
>>
>> MMUSIC *API*???
>
> Sorry, I was writing at a non-awake time.
> I meant the WebRTC API (Javascript, specified in the W3C WG).

OK. That makes more sense.

Well, there are two solutions to that:

- enhance the API to provide that capability, perhaps with a default
   to make it simple in common cases.

- limit WebRTC to follow a convention for which only supports a
   subset of what is possible in SDP.

IMO the first is the wiser choice.

In any case we are not defining this in SDP solely for WebRTC. We need 
to specify this in a way that works for everybody that wants to use it.

One way or another the SDP must define the SCTP association, and if it 
is to share a 5-tuple then the grouping to do that needs to be defined.
IIUC, the current bundle draft isn't capable of doing that.

	Thanks,
	Paul