[JDEV] Videoconferencing with jabber / Re:[speex-dev]Videoconferencing with speex and jabber
alon at sreu.com
Tue Nov 25 10:29:41 CST 2003
Ulrich B. Staudinger wrote:
> Am Di, 2003-11-25 um 15.00 schrieb Richard Dobson:
>>>>Just imagine you have ten people in a conference talking with each other
>>>>.... a peer would send ten streams to ten people, that's quite a waste
>>>>of bandwidth IMO.
>>>True, but you don't *have* to do it this way. Perhaps some sort of
>>>circular architecture -- put all the participants in a "ring" -- each
>>>participant sends a message only to the next one on the ring which will
>>>forward the message to the next and so on and so forth until the person
>>>who sits before the one who sent the message gets it.
>>>Anyway, P2P has lots of NAT-traveling problems.
>>Yep but there are ways around the NAT problem in the majority of cases as
>>already shown, but I do like this circular p2p architecture idea for
>>conferences, seems like quite a good solution if worked on to work out and
>>solve any problems.
> Token ring in voice is simply nonsense. What if one of the users has a
> slow machine? Every machine would have to 1) decode incoming data, 2)
> mix his own data 3) encode again.
> If we have 10 machines, and each machine takes 2ms to encode 10 ms of
> data, encoding 10 ms through the ten machines would take 20 ms (without
> packet travel time) for a frame to arrive at the first machine.
Actually I wasn't thinking about each user decoding/mixing/encoding, but
rather simply forwarding the raw packet to the next one in the ring.
Each endpoint has an incoming and outgoing queue -- when packets arrive
at the incoming queue they're forwarded to the outgoing queue to be sent
to the next point in the ring (without decoding), and also decoded
locally. When a point wants to send something it injects the encoded
packet to the outgoing queue.
Perhaps it is nonsense, I don't have any experience in such systems, the
idea just popped into my head while writing the email.
More information about the JDev