Re: [MMUSIC] SDPCapNeg, modified m-line issue

[Flemming Andreasen <[e-mail]>]

Christer Holmberg wrote:
> Hi,
>
>   
>>> I don't think having a 2nd offer/answer breaks the requirement (REQ-100). 
>>>       
>> Agreed (and this is also what we have in the current draft). 
>>
>>     
>>> The capabilities ARE negotiated within a single offer/answer. The second offer/answer is then used to "synchronize" 
>>> possible intermediates. And, if there are no intermediates, there is no need for a second offer/answer even with the 
>>> changes proposed by Ingemar.
>>>       
>> Agreed and again, this is also what is in the current draft, however I don't believe this is what Ingemar is proposing. 
>>     
>
> As I understand, Ingemar is proposing that the m= line is not changed in the answer. And, my point is that even in that case the capabilities WOULD still be negotiated in a single/offer answer. After all, the "final choise" is not indicated by the m= line, but by certain attributes.
>   
I'm not sure what you mean by "negotiating capabilities" so let me try 
and clarify: In the current draft, the offerer provides an actual 
configuration for one or more media streams, a set of capabilities, and 
one or more alternative potential configurations using those 
capabilities. The answerer in turn chooses one of the configurations 
(actual or potential) and then generates an answer indicating the now 
negotiated media stream based on that configuration. The answerer may 
also include additional capabilities in his answer (but doesn't have to 
do so).

As I understand Ingemar's proposal, he instead wants to have two O/A 
exchanges where in the first exchange, there is merely an exchange of 
capabilities, and in the second exchange, based on the capabilities 
exchanged, he then negotiates the actual configuration for the media 
stream. As I noted in a previous response, it is indeed possible to 
satisfy those semantics with the current draft, albeit with a few 
limitations on expressiveness. Those limitations however are lifted by 
use of the media capabilities extensions (latent configurations in 
particular) and hence we have a solution to the problem that was posed 
that 3GPP and others can use if they don't upgrade their intermediaries.

>   
>>> The big advantage is that intermediates not understanding SDPCapNeg will only see "normal" offer/answer transactions.
>>>       
>> Maybe so, but the media streams for sure will not reflect that "normal" offer/answer transaction and hence may use very 
>> different resources than the intermediary expected.
>>     
>
> That is true, but THAT can be fixed with a second offer/answer. The problem AT THE MOMENT is that it is difficult to say exactly what intermediates will do, what resources they will reserve (if they will reserve anything, or if they will release the whole stream/call).
>   
Agreed. Given that however, I have a hard time seeing why you think your 
proposal is any better since we both seem to agree that it is entirely 
speculative whether a particular intermediary will have problems with 
the current solution, the alternative Ingemar proposed, or both.

> Even with the current draft I think, in order to be on the safe side, that a UA should always send a second offer/answer, because the UA has no clue about whether possible intermediates "need" it or not.
>
>   
Right - and that is also what the draft specifies in Section 3.6.3.

>> As I noted previously, we discussed this option last year, but declined to pursue it (see below mentioned thread). Since 
>> the concerns you are raising here are essentially the same as were raised last year, and consensus back then was for the 
>> current solution, I think it's a bit late to try and change that at this point. 
>>     
>
> I am aware that this is coming late, and I appologize for that, but it's not the first time things are re-discussed. One difference is that now we also have the CS draft from Miguel, which allows the change of the c= line, which may make the whole thing even worse (see separate thread).
>   
Sure, and we can probably find loads of other parameters both new and in 
the future that compounds the problem. At the end of the day, the 
problem boils down to having an intermediary trying to do things with an 
SDP it only partially understands. This is a fundamental and unsolvable 
problem that applies to all intermediaries, whether they are NAT ALGs, 
Firewall, 3GPP P-CSCFs, PacketCable CMSes, etc.  I really don't see 
anything new in this discussion, and as I noted, we've been through it 
several times before, we designed the solution around the consensus 
decision, and I have furthermore outlined a solution based on the 
current mechanism that seems to address the problems you have described.

-- Flemming

> Regards,
>
> Christer
>
>
>
>
> 	 
> 	Regards,
> 	 
> 	Christer
> 	 
> 	 
>
> ________________________________
>
> 	From: [e-mail] [mailto:[e-mail]] On Behalf Of Flemming Andreasen
> 	Sent: 20. toukokuuta 2008 0:00
> 	To: Ingemar Johansson S
> 	Cc: [e-mail]
> 	Subject: Re: [MMUSIC] SDPCapNeg, modified m-line issue
> 	
> 	
>
>
> 	Ingemar Johansson S wrote: 
>
> 		Hi
> 		
> 		Thanks for the comments. To your questions/comments.
> 		
> 		Sofar e.g CSCF's are identified as nodes that may be problematic. In many cases it is likely that a B2BUA (e.g an SBC) may act as a "translator" (between e.g IMS and non-IMS) as it may anyway need to handle things such as encryption and transcoding but still a P-CSCF in the path may cause problems for different scenarios and may increase the number of interop scenarios and issues that must be dealt with. It is really difficult to say what might happen ( http://tools.ietf.org/html/draft-ietf-sipping-sip-offeranswer-08 does not consider unacceptable answers).
> 		
> 		It is true that some kind of a mandate for a 2nd offer/answer will very likely in the end lead to that media will not flow until the 2nd offer the earliest. Currently it is unclear if it will lead to increased setup times or if it is hidden by other signaling (e.g precondition)
> 		
> 		I tried to find some info about the modified "m=" line in the SDPCapNeg draft, more specifically if it is connected to a MUST. Sofar I have not found any such requirement. I may have missed it, please correct me.
> 		  
>
> 	It follows from the description in Section 3.6.2 (Generating the answer), and in particular the following excerpt: 
> 	
> 	Once the answerer has selected a valid and supported offered 
> 	   potential configuration for all of the media streams (or has fallen 
> 	   back to the actual configuration plus any added session attributes), 
> 	   the answerer MUST generate a valid answer SDP based on the selected 
> 	   potential configuration SDP, as "seen" by the answerer (see Section 
> 	   3.6.2.1. for examples).
> 	
> 	
>
> 		The only more specific info I found is in section 3.4.2 that says that the receiver of the answer must be able to handle the case that an intermediary may rewrite the "m=" line. 
> 		  
>
> 	I'm not sure where in Section 3.4.2 you see that. Answers are supposed to follow the rules in Section 3.6.2. and if an intermediary changes that, then things may or may not work (if it's a B2BUA, then obviously we are in a different situation). 
> 	
> 	
>
> 		This would then mean that the answer may leave the "m=" line unaltered ?. 
>
> 	No - see Section 3.6.2
> 	
>
> 		The offer when it receives the answer will then see that the "a=acfg" and the "m=" line does not correlate well and will issue another offer/answer round (probably as an UPDATE). There may occur some transmission of "faulty" media but as the receivers MAY discard this media this should be a minor issue (could be some early media security issues though).
> 		  
>
> 	Again, this is not what the draft specifies (see Section 3.6.3 as well)
> 	
> 	-- Flemming 
> 	
> 	
>
> 		Regards
> 		Ingemar
> 		
> 		
> 		
> 		  
>
> 			-----Original Message-----
> 			From: Flemming Andreasen [mailto:[e-mail]] 
> 			Sent: den 18 maj 2008 22:19
> 			To: Ingemar Johansson S
> 			Cc: [e-mail]
> 			Subject: Re: [MMUSIC] SDPCapNeg, modified m-line issue
> 			
> 			Hi Ingemar
> 			
> 			Sorry for the delayed response - please see below.
> 			
> 			Ingemar Johansson S wrote:
> 			    
>
> 				Hi
> 				
> 				As you may know, SDP Capability Negotiation Framework 
> 				      
>
> 			(SDPCapNeg) is gaining interest in the 3GPP community 
> 			especially in the SA4 and CT1 working groups. 
> 			    
>
> 				The current plan is to propose SDPCapNeg to be included as 
> 				      
>
> 			a part of the 3GPP standard.
> 			    
>
> 				There are however concerns about a potential 
> 				      
>
> 			interoperability issue with SDPCapNeg. 
> 			    
>
> 				Namely that e.g. the m= line in the (first) SDP answer is 
> 				      
>
> 			modified in such a way that intermediaries may reject the SDP 
> 			with unpredictable/unknown consequences (also mentioned in 
> 			3.12 in the SDPCapNeg draft).
> 			    
>
> 				An example is the SDP
> 				Offer:
> 				   m=audio 1234 RTP/AVP 97
> 				   a=fmtp..
> 				   a=rtpmap..
> 				   a=tcap:1 RTP/AVPF RTP/AVP
> 				   a=pcfg:1 t=1|2
> 				
> 				The answerer supports AVPF and returns
> 				Answer:
> 				   m=audio 5678 RTP/AVPF 97
> 				   a=fmtp..
> 				   a=rtpmap..
> 				   a=acfg:1 t=1
> 				
> 				An intermediate may accept the offer but reject the answer 
> 				      
>
> 			for some reason and as it is the answer that is rejected the 
> 			consequences are worse than if the offer would be rejected.
> 			    
>
> 				  
> 				      
>
> 			Is this a general concern or are you looking at a specific 
> 			intermediary (e.g. the P-CSCF) ?
> 			
> 			    
>
> 				So far this is only a problem related to the m= line but as 
> 				      
>
> 			the framework allows for extensions that may elevate this 
> 			problem even more.
> 			    
>
> 				  
> 				      
>
> 			Agreed.
> 			    
>
> 				Our proposal is therefore that the SDPCapNeg answer is only 
> 				      
>
> 			allowed to do modifications to lines in the "conventional" 
> 			SDP parts that are well known to work or supported by 
> 			"conventional" SDP offer/answer exchange (the definition here 
> 			is yet unclear). The answer SDP above would then look like 
> 			    
>
> 				   m=audio 5678 RTP/AVP 97
> 				   a=fmtp..
> 				   a=rtpmap..
> 				   a=acfg:1 t=1
> 				In this example the preferred configuration is only 
> 				      
>
> 			indicated by the a=acfg line.
> 			    
>
> 				  
> 				      
>
> 			Would the media stream(s) have been established with the 
> 			indicated answer at this point in time (A) or would you now 
> 			require a subsequent offer/answer exchange before the media 
> 			stream(s) can be considered functional (B) ?
> 			    
>
> 				This could of course make it more necessary to do a 2nd 
> 				      
>
> 			offer/answer exchange, but we believe that mandating a 2nd 
> 			offer answer will be needed even with the current solution to 
> 			make sure things will work.
> 			    
>
> 				One could rightly argue, as is also hinted in the SDPCapNeg 
> 				      
>
> 			draft, that intermediaries should be upgraded. Problem is 
> 			however that depending on product cycles and other issues 
> 			among vendors and operators, this may take time. Also it is 
> 			worth notice that 3GPP works based on the principle that 
> 			things should be backward compatible. Furthermore we believe 
> 			that solutions based on the "products should be upgraded" 
> 			principle would cause problems even in non-3GPP networks 
> 			since intermediaries may be found in many network deployments.
> 			    
>
> 				Comments and suggestions on this issue are welcome.
> 				  
> 				      
>
> 			Taking a step back, I believe the essence of your comments 
> 			are, that if we want to use SDP Capability Negotiation to 
> 			establish media streams, then we cannot do that in a single 
> 			offer/answer exchange. Instead, you want to have a mechanism 
> 			whereby we first exchange capabilities between the offerer 
> 			and the answerer (without establishing a stream) followed by 
> 			another exchange where we actually negotiate the media stream 
> 			parameters. In other words, the current single-roundtrip O/A 
> 			exchange provided by RFC 3264 and supported by SDPCapNeg is 
> 			replaced by a two-roundtrip solution.
> 			
> 			While I understand the concern you are trying to address, we 
> 			have been around this issue several times and the sdpcapneg 
> 			document reflects the consensus solution. To recap, the 
> 			requirements document
> 			(draft-ietf-mmusic-sdp-capability-negotiation-reqts-01) has 
> 			the following requirement:
> 			
> 			   REQ-100: The mechanism MUST work within the context of the
> 			   offer/answer model [RFC3264]. Specifically, it MUST be possible to
> 			   negotiate alternatives within a single offer/answer exchange.
> 			
> 			That particular requirement was discussed in the design team, 
> 			in the Prague IETF, and subsequently on the MMUSIC mailing 
> 			list (see "SDPCapNeg Issue #2: Answer not getting through 
> 			middle-boxes if transport protocol in answer differs from 
> 			offer" on 6/25/07) and consensus was for the mechanism 
> 			specified currently.
> 			
> 			Note however, that the SDPCapNeg framework as currently 
> 			defined can be used more or less the way you seem to prefer 
> 			by merely including capabilities in the offer without the 
> 			actual potential configuration attributes (which are used to 
> 			trigger the extended O/A exchange). The thing that would be 
> 			missing is expressing valid combinations of capabilites as 
> 			well as preferences, however this is part of what the media 
> 			capabilities extension document is addressing by introducing 
> 			the "latent configuration" attribute, which you could then use.
> 			
> 			In lieu of the above, past history on requiring multiple O/A 
> 			exchanges, the significant changes to the overall scheme your 
> 			suggestion implies, and the fact that we completed WGLC a 
> 			while ago, I don't think we should make this change.
> 			
> 			Regards
> 			
> 			-- Flemming
> 			
> 			    
>
> 				Regards
> 				Ingemar
> 				*******************************************
> 				Ingemar Johansson
> 				Senior Research Engineer, IETF "nethead" 
> 				EAB/TVP - Multimedia Technologies
> 				Ericsson Research Ericsson AB
> 				Box 920 S-971 28 Luleå, Sweden
> 				Tel: [phone]
> 				ECN: 850-43042
> 				ECC: 850-43074
> 				Mobile: [phone]
> 				*******************************************
> 				_______________________________________________
> 				mmusic mailing list
> 				[e-mail]
> 				https://www.ietf.org/mailman/listinfo/mmusic
> 				
> 				  
> 				      
>
> 		
> 		  
>
>
>   

_______________________________________________
mmusic mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/mmusic