Re: [MMUSIC] [AVTCORE] BUNDLE: a single stream with multiple MIDs?

[Miguel París Díaz <[e-mail]>]

Hello everyone,
firstable, thanks Jonathan to open this interesting discussion.

For me it is quite important not only having real time
applications/infrastructures working, but also in an efficient way, because
it could be the different between success or failure...
Why? Because the usage of technology is subject of business, where the
costs are essential.

Taking this into account and that in the real world:
  - For conferences with 3 or less participants (usually) the best approach
is doing p2p.
  - Conferences with 4 or more participants (usually) use a midlebox like
an SFU, etc.
  - Most conferences have less than 5/6 participants

In the Jonathan's case with a SFU:
  - Sending all streams, there would be nParticipants input streams and
nParticipants * nParticipants output streams. Total streams: (nParticipants
+ 1) * nParticipants
  - With his proposal, there would be nParticipants input streams and
(nParticipants -1) * nParticipants output streams. Total streams:
nParticipants * nParticipants, so we save nParticipants streams.

And the percentage of streams saved is: 1 - ( (nParticipants *
nParticipants) / (nParticipants + 1) * nParticipants)) = 1 - (nParticipants
/ (nParticipants + 1))

Assuming that in an SFU the cost of receiving and sending streams is almost
the same, with this approach we could save CPU and reduce the bandwidth
usage with conferences of:
  - 4 participants: 1 -4/5 = 1-0.8 = 20 %
  - 5 participatns = 1- 5/6 = 1 - 0.833 = 16.6 %
  - 6 participants = 1 - 6/7 = 1 - 0.857 = 14.3 %

We also should consider that using simulcast of SVC for video would reduce
this percentages.
But in conclusion, let's take into account the efficiency, not only for
that but in general please ;)

Related to that:

Then a=mid comes to the rescue and provides a per m-line unique
> identifier, leaving the door open for a future spec change allowing
> m-lines removal (have you ever seen how an SDP from a SFU with
> participants entering and leaving the room grows?).
>

To avoid huge SDP, there is draft proposing partial offers and answers:
https://tools.ietf.org/html/draft-roach-mmusic-pof-pan

Best regards!!

-- 
Miguel París Díaz
------------------------------------------------------------------------
Computer/Software engineer.
Researcher and architect in http://www.kurento.org
http://twitter.com/mparisdiaz
------------------------------------------------------------------------