Re: [MMUSIC] When a bundle offer is forked

[Martin Thomson <[e-mail]>]

The more I think about forking, the more I don't want to think about
it any more.

On 23 May 2013 10:34, Paul Kyzivat <[e-mail]> wrote:
> * if the bundle is accepted by multiple forks, does this form
>   one RTP session or several?

If the subjective view of the RTP session that the offerer holds is
bound to a 5-tuple (as opposed to just local addressing), then this
forms multiple sessions.

However...

Prior to getting answer(s), the offerer might be unable to tell that
there are multiple RTP sessions, or if it is just media arriving from
multiple source addresses.  This would only be a problem in the
absence of ICE and with multiple m-lines.

> * How does ICE work in this case? (Is that already covered by ICE,
>   since it mostly isn't specific to bundle?

I believe that ICE works in this case.  Each RTP session should have a
different set of ufrag/pwd for the peer.

I know that others have talked about the implications for the ICE
state machine for bundle, but that depends on ufrag/pwd.

> * If one fork adds/removes an m-line from the bundle, should that
>   be mirrored to the other forks?

I don't see how that has a significant impact.  Each fork could choose
different bundling configurations.  This is not significantly
different to other forking cases.  The offerer has to be prepared to
demux at the session layer first.