--- --- Multiparty Multimedia Session Control (MMUSIC) Working Group --- IETF#69 Meeting Minutes --- Tuesday, 24 July 2007 from 15:20-17:20 --- Wednesday, 25 July 2007 from 13:00-15:00 --- ============================================================ Reported by Jean-François Mulé based on notes from Colin Perkins and Jean-François Mulé The MMUSIC WG met twice at the 69th IETF. The meeting was chaired by Colin Perkins, Jörg Ott, and Jean-François Mulé. Both sessions were attended by some 100+ participants. 1/ Introduction and progress report The chairs presented the agenda and a brief status of the working group: - no RFCs were published between IETF#68 and IETF#69 - draft-ietf-mmusic-securityprecondition-04 was in IESG review at the time of IETF#69 and is in the RFC Editor queue as of August 21 2007 - draft-ietf-mmusic-sdpng-trans-04 is dead. For agenda details, see file named mmusic-0.pdf 2/ Source-Specific Media Attributes in SDP http://tools.ietf.org/html/draft-lennox-mmusic-sdp-source-attributes-01.txt Jonathan Lennox presented the source-specific media attributes draft. There was some discussion on whether there should be an SDP attribute for an agent to indicate that it can handle a maximum of N sources (slide 7). Some believed the mechanism is not required since in ASM, it's never been possible to control this and implementers need to realize that RTP is a group communication protocols; others raised the issue and complexity of dealing with errors and what to do if an agent can only support one source, which one it could/should select, etc.). Based on the above comments from Steve Casner, Roni Even, Colin Perkins and other participants, the consensus was that this functionality should not be added to the current draft. On the usages of source-specific fmtp (slide 8), Roni Even noted that the draft is unclear on which stream will be chosen if the remote agent does not support this attribute. Flemming commented that defining a means to indicate source-specific fmtp parameters at a media attribute level could lead to numerous offer/answer exchanges and could constitute a major change to the offer/answer model (since a=fmtp is negotiated in the offer/answer exchange). Flemming recommended that more general guidance be provided on how this interworks with offer/answer. The discussion ended without any clear consensus and more will follow on the list. John Elwell asked how this draft interacts with sip forking & multiple streams, and if it could help correlating the call legs. Christer Holmberg also wondered if it can help associate a media stream with a particular dialog. Jonathan Lennox responded that it may help but other mechanisms like ICE could be used. Jonathan Lennox would like to move quickly on the base spec due to the RTCP-SSM dependency. Colin Perkins asked the working group for the level of interest in this draft via a hum (consensus was "yes"). Subject to rechartering, there was consensus that this work should progress in MMUSIC. 3/ Media Description for IKE in SDP http://tools.ietf.org/html/draft-saito-mmusic-sdp-ike-01.txt Makoto Saito presented the draft update for IKE media description in SDP. The draft has been updated by Makoto Saito and Dan Wing and proposes to use the same concept as comedia-tls to exchange the fingerprint of the self-signed cert for IKE authentication. Eric Rescorla (EKR) said that there should be a means to indicate that IKE leads to IPsec. Dan Wing clarified that some semantics may be missing in the draft to indicate that the IKE negotiation of a security association will lead to an IPsec tunnel. EKR also noted that in the presence of NATs, it is not sufficient to punch holes for IKE using ICE since IPsec is a separate protocol. There were additional discussions on how an agent could signal that it wants to initiate an IKE negotiation. Colin Perkins noted that the use of "m=application 500 udp ike" which defines application/ike might be controversial. The chairs also noted that the SIPPING working group should also be aware and folks from SIPPING should be involved. It was agreed that the MMUSIC chairs will talk with the AD Cullen Jennings to figure out the best path forward so the authors can make progress. May be MMUSIC is the home for SDP issues, and other parts of the drafts can be homed elsewhere. 4/ Configuring DSCP of SDP Established Media Streams http://tools.ietf.org/html/draft-polk-mmusic-dscp-attribute-01.txt James Polk presented an update of the draft proposing to "configure" (offer or dynamically change) the DSCP markings of a session's media streams. Several comments were made on the open issues (slide 7): - Francois Audet first recommended to not to use DSCP code points but something that is more high level; Francois also added that while an SDP attribute may make sense for end-to-end, if the mechanism relies on an SBC or network middle box to work, it should use subscribe/notify rather than allowing SDP to be modified midcall - Dave Oran has several issues with the draft: one, the mechanism is not granular enough to specify which packet of a given media stream a certain DSCP is applied to (e.g. media packets of I and P frames of an MPEG2-TS with 2 different DSCP values). Secondly, there are concerns about rewriting DSCP across domains and the problem space may be bigger than the current draft. - James Polk responded that the scope could be expanded beyond the current "per-stream solution", the draft is open for suggestions on this - Hadriel Kaplan insisted that a B2BUA is a UA and can change SDP - Other comments followed from Jean-Francois Mule, Brian Stucker raising concerns about UAs across domains marking packets based on the remote UA's local DSCP policy, and whether a server has more of better knowledge than the UA. - James responded that a solution could be applicable to a private network where the above is known - John Elwell asked if the SIP Session Policy framework could help (James does not think so). - Dave Oran concluded the comments by raising the question of whether a more abstract mechanism could be proposed: signal "service X" instead of "DSCP X" which implements that service). It was proposed to organize a meeting while at IETF or organize a conference call after IETF in order to discuss the problem statement, scope of applicability and set of issues. 5/ SDP Offer/Answer Mechanism for File Transfer http://tools.ietf.org/html/draft-ietf-mmusic-file-transfer-mech-03.txt Miguel Garcia presented the recent changes and open issues on the SDP mechanism for file transfer. Related to a bullet point referencing a recent change on hash algorithms (slide 6), Jonathan Rosenberg asked why multiple hash algorithms are needed and EKR commented on the purpose of the hash (integrity check vs. security feature). EKR recommended using something other than a hash like a specific CRC if it's a CRC. There were no comments from the working group on the proposal to refer to MSRP and make message/cpim is mandatory to implement for the message/cpim wrapper. On the open issue of aborting existing file transfers, based on questions from Jonathan Rosenberg, Gonzalo Camarillo clarified the use case description. One of the requirements is to have a mechanism to restart a transfer, including an explicit signal to start the file transfer over in a re-invite and distinguish an explicit restart request from a message repeat. Based on comments from Jonathan Rosenberg and Joerg Ott, it is proposed to add some kind of transfer identifier field which could for example change if the user wants to abort. The authors will revised the draft which should then be pretty close to WGLC. 6/ Multiple Packetization Times in SDP http://tools.ietf.org/html/draft-garcia-mmusic-multiple-ptimes-problem-00.txt This draft describes a problem statement for multiple packetization times in SDP. Colin Perkins and Joerg Ott asked why this is needed. Hadriel Kaplan, Robert Sparks, Jean-Francois Mule provided examples including express requests from the implementers community of media gateways, limitations in DSP silicon some vendors have, accurate QoS budget calculations in some networks. Robert Sparks mentioned that in SIP-it, while issues related to the lack of such a mechanism are not widely spread (below 10%), some implementations break at every SIP-it so a clarification for implementers would be good. There were divergences in what should be done about this problem, either produce a fix or recommendation for so-called "broken implementations". Colin Perkins stated that we should not assume the outcome and that telling folks that their implementation is broken is a possible outcome (it may not be the consensus but a possible outcome). It was concluded that more input is needed on this topic. 7/ Signaling media decoding dependency in SDP http://tools.ietf.org/html/draft-schierl-mmusic-layered-codec-04.txt Thomas Schierl presented the updates done in the draft addressing how media decoding dependencies could be expressed in SDP. This is in part motivated by RTP session multiplexing of layered media or multi description coding. There was some debate on where those media decoding dependencies fit (media or session level attribute). Roni Even asked why dependencies are done at the media attribute level and not at the session level. Magnus Westerlund justified its definition as a media attribute since it is the media stream that has dependency. Another discussion was about the scope of the document and the choice of the grouping framework as part of the solution. Flemming Andreasen commented that it is not clear that grouping is the right framework: how can an agent express an offer with 2 alternative video codecs? Thomas proposed to use different groups to express this Magnus and Flemming agreed that there are cases where this does not work and get complex. Magnus concluded this discussion by proposing to limit the scope of this document per payload type; this could solved in a different draft. This scope question should be confirmed on the list. Pending rechartering approval, there was consensus for this draft to become a wg item. 8/ RTSP 2.0 http://tools.ietf.org/html/draft-ietf-mmusic-rfc2326bis-15.txt See slides for details. Dave Oran pointed to various project initiatives and SDOs modifying RTSP: ITU, ATIS, 3GPP, DVB, etc. Jean-Francois Mule indicated he had calls with the editor of the RTSP requirements in DVB trying to get them to contribute to IETF. Dave Oran, IAB requested the chairs send liaison statements to those various organizations to invite people to bring their work to IETF on 2.0 RTSP. 9/ RTSP 2.0 and NAT Traversal http://tools.ietf.org/html/draft-ietf-mmusic-rtsp-nat-evaluation-00.txt Magnus Westerlund presented the work initiated on RTSP 2.0 and NAT traversal by outlining that the problem to solve is the NAT traversal of the media stream from the RTSP server to the RTSP client. As for the solution, comments were made in favor of continuing with the previous consensus that the solution relies on ICE (slide 6). 10/ SDP Capability Negotiation http://tools.ietf.org/html/draft-ietf-mmusic-sdp-capability-negotiation-06.txt Flemming Andreasen presented the latest drafts on SDP capneg (draft-ietf-mmusic-sdp-capability-negotiation-06, draft-andreasen-mmusic-sdpcapneg-att-del-00 and draft-ietf-mmusic-sdp-capability-negotiation-reqts-01) - see slides for details. The open issues in the core specification (draft-ietf-mmusic-sdp-capability-negotiation-06) were discussed first: - 1) Lack of support for bandwidth capabilities (slide 7): The consensus was to continue discussing the requirements and examples on the list. Magnus agreed to write up something (an example?) on how to use the for codec and bw modifiers for negotiations in the core spec. - 2a) Lack of support for media capability negotiations (slide 8): Magnus stressed the importance of this but Flemming and Ron commented that there is no requirement to justify this is in the core. To be added, there should be a requirement for it. - 2b) Session level coordination between potential configs (slides 9 & 10): There were no comments from the wg which suggests that this would be left out of the core unless someone brings a strong use case scenario. - 2c) Media capabilities negotiation, in or out of core (slide 11)? Consensus: leave it out unless someone comes up with a strong example and case in the near term (within 2 weeks or so). Some of the discussion items are captured below. Jonathan Rosenberg, Roni Even and Flemming made comments towards leaving media capability negotiation out of the core. Jonathan indicated that we should leave it out in the interest of making progress for the negotiation of RTP | SRTP ("SRTP problem"). Magnus agreed but had concerns about forward compatibility and not having enough flexibility for later (the tcap param is not generic enough to support future extensions). Steve Casner commented that the main problem with SDP is that attributes you don't understand are ignored. This makes extensions hard, because you can't tell if they're being used or not. If you've addressed that, the problem is a lot easier to solve. - 3) Usage of Truncated Syntax as capability definitions (slide 12) The group's consensus was to remove this functionality. Jonathan voted for "less is more". Joerg Ott raised the issue that you may end up with arbitrarily truncated attributes for which the meaning is unclear. - 4) Media Before Answer (slide 13): Consensus: leave as-is - 5) Obsoleting RFC 3407 (slide 14): Based on input from Francois Audet, and Flemming, and the fact that RFC 3407 is deployed in some networks, the consensus was to not obsolete RFC 3407 (option 3) in slide 14) This is document is getting ready for Working Group Last Call unless Magnus or anyone else comes up with something new or major that should be addressed. 11/ ICE TCP http://tools.ietf.org/html/draft-ietf-mmusic-ice-tcp-04.txt Jonathan Rosenberg presented the changes and open issues in ICE-TCP. - Open issue #1: DTLS-SRTP EKR indicated that there is no technical reason why this won't operate correctly. Francois Audet agreed that this is not as bad as it sounds. Dave Oran is ok with the recommendation but noted that what is deployed in the field is RTP over TCP and problems will arise when the negotiation of RTP over TCP is different from DTLS-SRTP over TCP. Colin Perkins recommended using DTLS-SRTP over TCP instead of RTP over TLS which should be avoided. Jonathan agreed. We had general consensus to run DTLS-SRTP over TCP rather than RTP over TLS. Steve Casner proposed that the document should have a paragraph to capture why RTP over TLS should not be done. - Open issue #2: S-O candidate from TURN server Phil Matthews expressed concerns that we do not fully understand the combinations of NAT cases that are eliminated if this is removed and indicated that for p2psip in particular, having this functionality may be an interesting case. Cullen outlined a really unlikely use case (2 peers behind fully symmetric NAT and one peer uses a TURN server behind a NAT?). Jonathan agreed to reconsider this whole case in the light of p2psip which seems to be a valid case. More considerations is required given the implementation impact -> more discussion to the list is expected on this. - Open issue #3: TURN vs. RTP shim (RFC4571) There was general agreement to keep using the RTP shim (RFC4571) as current defined in the draft. This consensus was based on comments from Colin (given what's deployed, it is good to stick with 4571), Jonathan Rosenberg and Rohan Mahy. ICE-TCP needs more reviews pending some updates. The meeting concluded. > end.