CURRENT_MEETING_REPORT_ Reported by Abel Weinrib/Bellcore Minutes of the Multiparty Multimedia Session Control Working Group (MMUSIC) An on-line copy of the minutes and the accompanying slides may be found in the directory venera.isi.edu:confctrl/minutes as files ietf.11.93 and slides.[a-d].11.93.ps. The MMUSIC Working Group met for two sessions at the Houston IETF meeting. The first day was dedicated to a short overview of the goals and context for the working group and a presentation of an algorithm and framework for managing shared session state. The second meeting focused on preliminary ideas as to what might comprise shared session state for a couple of different session types, and three short presentations on related work. Overview and Framework Abel Weinrib presented an overview of the goals of the MMUSIC Working Group and discussed the framework for the work. This presentation was basically a review of the work of previous working group meetings; refer to the minutes of those meetings available from the confctrl archives for more detail. In setting the context for the next two talks, a distinction was made between the ``agreement algorithm'' and the ``session control protocol.'' The agreement algorithm supports generic control of group membership and enforces correctness and other policies on state shared among the members. This agreement layer ``understands'' membership and policies, but views the rest of the domain-specific session state as opaque. The session control protocol understands the domain-specific session state, using the services of the agreement protocol to manage the state shared among the members. The session protocol may also use other services in addition to the agreement services, such as services that support soft state sharing and recovery. Issues that were raised during discussion: o Where should a ``session manager'' that terminates a session control protocol reside? Various alternatives are on a workstation (as shown in the framework slide) for one or multiple users, or one per domain that could act as a demultiplexing agent by passing on session control messages for users in that domain to the appropriate place. The second alternative may provide hooks for supporting user mobility and may deal well with security firewalls. o Should floor control be done through the session control protocol or through some other mechanism? o Should policies be chosen from a predefined set, or should they be defined in all of their generality by each application? This has implications on interoperability and the complexity of the applications. o In the framework, resource reservation is separate from session management. The session control protocol is used to propagate a shared view of the state, which includes descriptions of the media streams required by a conference. An Algorithm for Managing Shared Teleconferencing State Scott Shenker described some preliminary ideas being developed for expressing policies about how session state can be changed and the degree to which members agree on their views of the state. Policy can be expressed along three dimensions: voting policies, consistency policies, and initiator policies. Voting policy defines which members must agree for a state change to take place. Consistency policies describe how the state seen by different members may differ. Initiator policies set which members may initiate changes to the state. The policy framework provides the vocabulary for concretely describing various session styles. He then presented an algorithm that supports operations on the shared state while enforcing the policies associated with the session. These operations might be adding a member, changing the policies themselves, or modifying some other domain specific state variable such as an encryption key. The basic mechanism is a group agreement algorithm based on a two-phase commit procedure or correctness. For additional information on this work there is a rough draft document in the confctrl archives in docs/agree.ps. Notice of the availability of more complete drafts of the document will be sent to the confctrl mailing list. Some points raised in the discussion during and following the talk: o It was observed that some members of a session may be programs running on computers. The fall-back position of always allowing members to leave a corrupted session may be less useful than for human members who can more easily detect the corruption. o Critical and non-critical membership allows there to be a core group of members that control the conference and a potentially much larger set of members that can more easily enter and leave. o This talk is about agreement, not negotiation. The distinction is that there is no support for multiple rounds of proposals and counter-proposals. This could be future work, or could be done at the application level building on top of the basic agreement service. Session Control Above the Agreement Protocol Eve Schooler's talk was devoted to the interpretation and usage of the agreement protocol for teleconference session control. Discussion attempted to place the agreement protocol in the context of a traditional protocol stack and to hint at implementation concerns. Examples were given for generic and domain-specific session operations, as well as for the array of potentially interesting state attributes (session-wide, membership-related, or media- and policy-specific). To illustrate the range of sessions that can be constructed from different sets of policies, two example paradigms were presented; one for an open hailing-channel session with little coordination among members, and another for a minimal invitation-only session. The second half of the presentation focused on several open issues: Tradeoffs between different end-system organizations, addressing issues related to the use of unicast and multicast and to the interaction of media agents and session agents, and alternate techniques for user rendezvous that resemble what is currently in place on the MBone for session directories. For additional information on this work, there is a very rough draft document in the confctrl archives in docs/usage.txt. Notice of the availability of more complete drafts of the document will be sent to the confctrl mailing list. Some points raised in the discussion: o Issues of media typing and the addressing of media agents are related to problems that need to be solved for WWW as well as XMosiac naming and MIME mailcap media descriptions. o It would be nice if session control did not assume that the media used by the conference is necessarily carried over an IP network. Consensus and Control in Wide-Area Communication Bala Rajagopalan briefly presented his work on agreement and control of group membership in wide area communications. He also handed out a paper that presents his model and algorithm in more detail; contact him via email for a copy of his paper. The model allows a group to (eventually) come to consensus on its membership in the presence of unreliable message delivery. His algorithm uses wide area multicast and a coordinator for each partition's ``view'' of the membership state. Operations on groups include join, leave, delete, reform, merge. One underlying assumption of this work that led to some heated discussion during the meeting is that connectivity is transitive, meaning that if A is connected to B and B is connected to C, then A is connected to C; this assumption may break down during certain failure scenarios in the Internet. This work appears to be relevant to the concerns of the MMUSIC Working Group. More effort is required to understand how and where it might fit into the MMUSIC charter. RTCP Implications for MMUSIC Steve Casner discussed the relationship of RTCP, the ``real time control protocol'' defined by the Audio/Video Transport Working Group (AVT), to the MMUSIC Working Group effort. RTCP is separate from the RTP protocol (which supports transport of time-critical media streams) and may in the future be replaced by a higher level control protocol, such as the MMUSIC session control protocol. In particular, he described the functions that RTCP currently provides, and discussed other functions that would be useful in supporting an application such as multimedia teleconferencing (see the slides). He concluded that it may make sense to use some part of the RTCP in conjunction with a higher level control protocol. Session Control Work at BBN Julio Escobar presented a list of relevant work at BBN that is addressing similar issues to the MMUSIC Working Group. He mentioned Chip Elliott's work on the ``sticky'' protocol (Chip had actually presented this work at an earlier MMUSIC/CONFCTRL BOF), Lou Berger's simulation exercise management tool, and Walter Milliken's work on resource coordination objects. Julio promised to send additional information on this work to the confctrl mailing list (which he has done). Attendees Lou Berger lberger@bbn.com David Borman dab@cray.com Stephen Casner casner@isi.edu Ping Chen ping@ping2.aux.apple.com George Clapp clapp@ameris.ameritech.com Steve DeJarnett steve@ibmpa.awdpa.ibm.com David Dubois dad@pacersoft.com Ed Ellesson ellesson@vnet.ibm.com Julio Escobar jescobar@bbn.com William Fenner fenner@cmf.nrl.navy.mil James Fielding jamesf@arl.army.mil Ron Frederick frederick@parc.xerox.com Atanu Ghosh atanu@cs.ucl.ac.uk Fengmin Gong gong@concert.net John Hanratty jhanratty@agile.com Ken Hayward Ken.Hayward@bnr.ca Van Jacobson van@ee.lbl.gov Yasuhiro Katsube katsube@mail.bellcore.com Charley Kline cvk@uiuc.edu Jim Knowles jknowles@binky.arc.nasa.gov Ted Kuo tik@vnet.ibm.com Paul Lambert paul_lambert@email.mot.com Mark Laubach laubach@hpl.hp.com Jim Martin jim@noc.rutgers.edu Thomas Maslen maslen@eng.sun.com Donald Merritt don@arl.army.mil Karen O'Donoghue kodonog@relay.nswc.navy.mil Laura Pate pate@gateway.mitre.org J. Mark Pullen mpullen@cs.gmu.edu Bala Rajagopalan braja@qsun.att.com Steven Richardson sjr@merit.edu Eve Schooler schooler@isi.edu Henning Schulzrinne hgs@research.att.com Scott Shenker shenker@parc.xerox.com Michael Speer michael.speer@sun.com John Stewart jstewart@cnri.reston.va.us Daniel Swinehart swinehart.parc@xerox.com Matsuaki Terada tera@sdl.hitachi.co.jp Claudio Topolcic topolcic@cnri.reston.va.us Abel Weinrib abel@bellcore.com Taehwan Weon weon@cosmos.kaist.ac.kr John Wroclawski jtw@lcs.mit.edu Shinichi Yoshida yoshida@sumitomo.com Lixia Zhang lixia@parc.xerox.com Weiping Zhao zhao@nacsis.ac.jp