Re: [pntaw] Real-time media over TCP

[Sergio Garcia Murillo <[e-mail]>]

+1,

IMO 6) will only have some benefits in the case one of the endpoint been 
a gateway, but you would still have to deal with 5), so you won't avoid 
needing a TURN server (in any of its flavors ;)

BR
Sergio
El 14/10/2013 17:08, Bernard Aboba escribió:
> Agree on Markus's summary. I am also not convinced on #6 being a MUST because so far there hasn't been a compelling use case.
>
>> On Oct 14, 2013, at 7:42 AM, "Hutton, Andrew" <[e-mail]> wrote:
>>
>> On: 11 October 2013 13:22  [e-mail] Wrote
>>
>>
>>> So, what is the overall set of traversal / connectivity / address
>>> gathering mechanisms the browser should support that you are
>>> advocating? Mine is:
>>>
>>> 1.) UDP host candidates (native interface) - MUST
>>> 2.)UDP server-reflexive candidates (STUN) - MUST
>>> 3.) UDP relay candidates via UDP (TURN) - MUST
>>> 4.) UDP relay candidates via TCP (TURN) - MUST
>>> 5.) UDP relay candidates via some HTTP/TLS compatible transport (TURN)
>>> - MUST/SHOULD
>>>    (TLS via HTTP CONNET or TURN over WebSockets have been proposed.)
>>> 6.) TCP based candidates (ICE-TCP) - MAY/SHOULD
>>>
>>> I believe 1.), 2.) and 3.) we all have consensus on, probably even on
>>> 4.). What we are discussing on this list is whether and how to do 5.),
>>> and how important 6.) is.
>>>
>>> My opinion is that 5.) becomes necessary due to the HTTP-limited
>>> environments you mention above. So the communication needs to be
>>> compliant with HTTP and its upgrade mechanisms. 6.) I see as an
>>> optimization to 4.) in cases where you can't do UDP but can establish
>>> outgoing TCP connections and the other peer can actually receive them.
>>> But that optimization does not hold very often, since most of the time
>>> browsers are in networks where they can't receive TCP connections from
>>> the "outside world". So that's why I see it as MAY or SHOULD. You may
>>> argue that often browsers may talk to gateways or central points that
>>> are reachable by TCP, and in those scenarios ICE-TCP might make sense.
>> Totally agree with the nice summary above and agree that 5) is necessary we have use cases in draft-ietf-rtcweb-use-cases-and-requirements that say we do. We need some mechanism that is HTTP compatible and we have these two options available and to me there is very little difference between them it is just a matter of whether the use of the additional websockets layers give us any benefit. I am not convinced that 6) is needed but might be some corner cases it satisfies.
>>
>> Andy
>> _______________________________________________
>> pntaw mailing list
>> [e-mail]
>> https://www.ietf.org/mailman/listinfo/pntaw
> _______________________________________________
> pntaw mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/pntaw