[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Simple] COMEDIA vs MSRP Relays: Shared Connections



On Jul 27, 2009, at 3:27 PM, Christer Holmberg wrote:


Hi,

More on this.

I'm not sure I follow you - Comedia has a "connection:existing"
attribute setting to re-use the same connection (e.g., in a new
offer).  Or do you mean of two separate media m= lines?

You are right of course. Brain fart on my part. However, the draft
should probably mention how to apply this--the only reference I see
to
connection:existing is in the ICE section.

Wait, I remember why I brought this up--if a peer is behind an MSRP
relay, how is it to know whether a reuseable connection already exists
downstream of that relay?

Even if it knows, how would it tell the relay whether to actually reuse
that connection? I guess that would require some
new information element.

We also need to think about the meaning of a=connection.

According to my understanding (I guess we can verify that with the
authors) of comedia, a=connection is not used to indicate that one can
use "any existing TCP connection" towards a specific destination.

My understanding is that a=connection is used during a session
modification, to indicate whether the existing TCP connection for that
specific session should be re-used (if possible) or not. So, for a
specific session, the change of address would always trigger a new TCP
connection - even if there exist TCP connections to the same address for
OTHER sessions.

Or?

Note, however, that according to section 4.1.3, connection re-use is
not used for initial offer/answers, only for re-
negotiations.


My read of COMEDIA agrees with yours. It's not clear to me if the connection attribute applies at all--RFC4975 says if there is already an connection associated with the same host, port, and URI scheme, it SHOULD reuse it. COMEDIA doesn't appear to have been written with that sort of thing in mind. I am leaning (not strongly) towards suggesting we ignore the COMEDIA connection attribute and just do what an MSRP client would normally do, at least for this purpose.


So, for session establishment, there is currently no difference between ACM and 4975 - if there is a TCP connection to a specific address, then
use it.

For 4975, I assume the same applies for session modifications
(eventhough 4975 doesn't explicitly says much about session
modifications)...

You are right, it doesn't say much. I agree that the same would apply here. Also, I think all of this so far is the same with or without relays.


ACM currently specifies the usage of a=connection for session
modifications, in accordance with comedia.

So, IF we a=connection to work with relays, I guess we would need to
provide the information to the relay inside the MSRP message.

I'm not sure I follow--what would you conceptually be telling the relay?


Again, I will add something about this to my presentation.

Regards,

Christer


-----Original Message-----
From: simple-bounces at ietf.org [mailto:simple-bounces at ietf.org] On
Behalf Of Ben Campbell

How does the COMEDIA negotiation interact with the TCP-connection-
sharing features of MSRP? We negotiate sessions in the SDP offer/
answer, not connections per se. COMEDIA implicitly assumes a one- to-

one correspondence between these, but MSRP explicitly allows
multiple sessions to use the same TCP connection.

I can see two approaches. One would be that the COMEDIA attributes
are only relevant if you have to create a new connection. If a
reusable connection already exists, you just use it. Alternatively,
we could state that a connection is only reuseable if it's direction

matches the COMEDIA attributes, and if devices wanted to reuse an
existing connection, they would either omit COMEDIA entirely, or
choose COMEDIA attributes that match the connection.

Neither of those approaches seems obviously wrong to me. I suspect
there may be some traps here, though--we should analyze this pretty
closely.

_______________________________________________
Simple mailing list
Simple at ietf.org
https://www.ietf.org/mailman/listinfo/simple

_______________________________________________
Simple mailing list
Simple at ietf.org
https://www.ietf.org/mailman/listinfo/simple

_______________________________________________
Simple mailing list
Simple at ietf.org
https://www.ietf.org/mailman/listinfo/simple
_______________________________________________
Simple mailing list
Simple at ietf.org
https://www.ietf.org/mailman/listinfo/simple