Re: [MMUSIC] draft-holmberg-mmusic-t140-usage-data-channel - multi-party

Hi,

>As I just replied to Christer, I think this discussion ought to pertain to a different document that we don't have yet.
>
>Meanwhile, having a channel per stream is hopefully sufficient while exploring these other things.

The pull request contains the following new text:

"If an implementation needs to support multi-party scenarios, implementations need
to support multiple simultaneous T.140 data channels, one for each remote party. At
the time of writing this document, this is true even in scenarios where each participants
communicate via a centralized conference server. The reason is that, unlike RTP media,
WebRTC data channels and the T.140 protocol do not support the indication of the source
of T.140 data. The SDP dcmap attribute label attribute parameter (<xref target="sec.sdp.dcmap"/>)
can be used to provide additional information about each T.140 data channel, and help
implementations to distinguish between them."

Regards,

Christer

	Thanks,
	Paul

On 8/28/19 2:26 PM, Gunnar Hellström wrote:
> Hi Paul, see comment at the end,
> 
> Den 2019-08-28 kl. 16:49, skrev Paul Kyzivat:
>> On 8/27/19 2:57 PM, Gunnar Hellström wrote:
>>> Hi Paul,
>>>
>>> Please see inline,
>>>
>>> Den 2019-08-27 kl. 20:39, skrev Paul Kyzivat:
>>>> On 8/26/19 3:59 PM, Gunnar Hellström wrote:
>>>>
>>>>> 4. A multi-party server S, combining a number of sources into one 
>>>>> call to a participant A, with real-time text from each other 
>>>>> participant (B,C,...) communicated in just one T140 data channel 
>>>>> between S and A. There is a need to indicate source for each 
>>>>> T140block sent to A. We currently have no way specified for that.
>>>>> An extension of T.140 could do it.
>>>>
>>>> Is there a reason to ever do this?
>>>>
>>>> In audio and video the "mixing" actually does something 
>>>> irreversible. And it takes real work to do it, and doing so reduces 
>>>> the bandwith considerably over what is required to transmit the 
>>>> individual streams.
>>>>
>>>> For RTT none of that is true. There is very little impact in 
>>>> bandwidth or processing in transmitting them all as separate channels.
>>>>
>>>> So why not just say "don't do that"?
>>>
>>> Yes, interesting and realistic thought. It would likely be the best 
>>> choice for many practical cases.
>>>
>>> I am not sure however how it will work with a huge conference with 
>>> hundreds of participants, and some of them occasionally asking for 
>>> the floor and send a bit of RTT text. A server using one data 
>>> channel per participant would either be required to establish an 
>>> enormous amount of T140 data channels being prepared to send what 
>>> has been received, or take the effort to establish a new T140 data 
>>> channel to all users at the moment a user gets the floor, and then 
>>> possibly close it again.
>>>
>>> What do you think about that case?
>>
>> Initially each user would only have one channel. Then he gets another 
>> one added each time there is a new speaker who hasn't spoken before. 
>> I don't know if there would be formal floor control, or if everyone 
>> would always be allowed to talk. Formal floor control would provide 
>> an early hint to establish the new channels, possibly preventing delays.
>> But adding a channel isn't an expensive operation and delaying 
>> messages until it is done shouldn't be a problem.
>>
>> But is this a realistic issue? Do RTT conferences with hundreds of 
>> speakers happen in practice?
> 
> It is realistic to the same degree as use of video and audio in 
> multi-party multi-media conferences.
> 
> It is not realistic to let hundreds of participants talk 
> simultaneously, but one or a very low number. It is realistic to let 
> one at a time have the floor and speak, sending the audio to all the 
> others, and have the opportunity to hand over the floor to someone 
> else with a lose or strict protocol.
> 
> There is very little use for transmitting live video from hundreds of 
> participants simultaneously, but it is realistic to show a low number 
> of users and be prepared to switch who are seen.
> 
> Similarly it is not realistic to let hundreds of participants present 
> new RTT simultaneously, but one or a very low number. It is realistic 
> to let a few at a time transmit, sending the RTT to all the others. 
> Anyone should have the opportunity to transmit RTT, controlled by 
> either no protocol or a lose or a strict protocol.
> 
> RTT is sadly often characterized as media for accessibility and 
> thereby expected to be used only in special cases. This is wrong from 
> two points of view:
> 
> 1: If used for accessibility, in cases when one or some users have no 
> or little use of audio or video for language communication because of 
> a disability, then it is very important that all other participants 
> can use the same media, so that communication can go directly to or 
> from the
> one(s) who need RTT. This is the whole idea of accessibility, that 
> users are not forced to limited technology corners, but are able to 
> participate anywhere with the same technology as others and have full 
> and equal participation. (sometimes a translating service between 
> different modalities are still needed so that everybody are enabled to 
> use the modality they prefer)
> 
> 2. In any multimedia conference there appears reasons to communicate 
> something in text. The three real-time media: audio, video and RTT are 
> needed together for alternating or simultaneous use. Talk, show and 
> write for a complete and efficient communication. In that scenario, 
> text messaging is a slow tool, causing delays and risk for losing the 
> interest from  viewers. When anyone starts texting RTT, it is possible 
> to start following the thoughts as they are expressed in text 
> immediately from start, while for messaging the typing user needs to 
> complete the message and risking to enter the message far too late to 
> be valid in the discussion.
> 
> 3. Automatic subtitling by speech-to-text services and even language 
> translation before presentation would be useless without a form of RTT.
> It is now realistic to use such services to enhance multimedia 
> conferences and make the contents available for both people with other 
> languages, people in noisy areas and for deaf or hard-of-hearing users.
> This is a rapidly increasing use. Also for this case it is realistic 
> only with a few actively transmitting users, while the result can be 
> if interest to distribute to many.
> 
> This said, I anyway agree that the usage that will develop most 
> rapidly will likely be for meetings with not more than five 
> participants having audio, video and RTT available and most commonly 
> only three taking turns on sending RTT. Some of these calls will be in 
> emergency services and interpreting (or relay) services, while there 
> is also a need for user-to-user conferences.
> 
> I would also like to know which MCU model is most realistic for these 
> applications. Is it the one with separate text channels per active 
> source opened and closed when there is a need, or is it the single 
> text channel with in-line identification of the source for each new 
> data channel message?
> 
> 
> Regards
> 
> Gunnar
> 
>>
>>
>>     Thanks,
>>     Paul
>>
>> _______________________________________________
>> mmusic mailing list
>> mmusic@ietf.org
>> https://www.ietf.org/mailman/listinfo/mmusic
> 
> 

_______________________________________________
mmusic mailing list
mmusic@ietf.org
https://www.ietf.org/mailman/listinfo/mmusic