Internet Info 1994 March/Internet Info CD-ROM (Walnut Creek) (March 1994).iso/inet/ietf/93mar/confctrl-minutes-93mar.txt

Text File | 1993-05-10 | 16KB | 405 lines

CURRENT_MEETING_REPORT_


Reported by Eve Schooler/Information Sciences Institute

Minutes of the Conferencing Control BOF (CONFCTRL)

These Minutes were prepared by Eve Schooler from notes provided by Abel
Weinrib of Bellcore and Deborah Estrin of USC/ISI.

Introduction and Presentations

Two CONFCTRL sessions were held at the Columbus IETF. The first meeting
was used to provide an overview of Conference Control efforts both
within and outside of the IETF. Inside the IETF, the CONFCTRL Group was
spawned by the Remote Conferencing Architecture BOF (REMCONF). Outside
the IETF, interest in conference control, sometimes referred to as
connection management, has been ongoing for some time.  Thus far, the
CONFCTRL mailing list has collected a sizable bibliography containing
references to many of the early and ongoing research projects in this
area.

Most of the first session was used for presentations on different
CONFCTRL schemes.  The intent of the presentations was to flesh out
design assumptions, tradeoffs, complexity, scalability, etc.  The
systems were classified according to several parameters:  whether they
(1) concentrate more on groupware conferencing (shared editors,
whiteboards) than on real-time audio/video conferencing, (2) provide
session control of packet-based real-time media versus analog real-time
media, (3) rely on centralized versus distributed session management,
and/or (4) observe loose versus tight session control.


   o Ruth Lang of SRI reported on the Collaborative Environment for
     Concurrent Engineering Design (CECED).

   o Abel Weinrib of Bellcore focused on the session elements and
     functions supported by Bellcore's Touring Machine.

   o Hans Eriksson of SICS discussed the CoDesk architecture.

   o Don Hoffman of Sun Microsystems outlined the model used for the
     COCO project.

   o Chip Elliott of BBN presented his work on VideoTeam and the Sticky
     CONFCTRL protocol on which it relies.

   o Lakshman Krishnamurthy of University of Kentucky summarized the
     versatile Multi-flow Conversation Protocol.

   o Eve Schooler of ISI gave an overview of the MMCC tool and its
     Conference Control Protocol.

                                   1





   o Thierry Turletti's ivs program was discussed as a contrasting
     example that uses loose-style session management.


Synthesis of CONFCTRL approaches

The second session was used to identify pervasive CONFCTRL themes, and
to question the applicability of the various solutions to the Internet.
The main objective was to narrow the scope of the problem en route to
the design of a generic CONFCTRL protocol.  Observations were culled not
only from the presentations at the IETF but also from templates that
were filled out prior to the meeting.  The templates included Dave
Lewis' write up of the UCL PREPARE project, a description of the ZAPT
project by Joe Touch of ISI, a contribution from Jack Jansen of CWI
about the Meeting project, and Fengmin Gong's template on the MCNC
CONCERT Video Network Migration effort.

Of particular interest were implementors' comments about the aspects of
their approaches which were hard, easy, or warranted change.  Except for
a lone comment about the ease of implementation of floor control, there
were several recurrent themes regarding implementation difficulties:


   o It is difficult to design a CONFCTRL protocol that balances
     simplicity with a high degree of semantic flexibility, e.g., Jansen
     of CWI concluded that different conferencing styles require
     entirely separate CONFCTRL protocols.

   o A distributed model comes with distributed system complexities:

      -  Support for causality of multiway message exchanges.
      -  Recovery from temporary network failures.
      -  Propagation of consistent state information.

     The solutions proved to be cumbersome, unexpectedly hard and often
     times ``tricky''.

   o The underlying transport (that carries session control information)
     comes at a price, e.g., the overhead of one RPC implementation led
     the PREPARE project to shift to a different, lighter
     implementation.

   o There is room for improved media integration, e.g., asymmetric
     flows are difficult to characterize at setup, there is a need for
     more powerful control over presentation of media streams.


Most experimental systems either are or began as LAN-based conferencing
systems.  However, it is clear that many, if not all, are aiming for WAN
operation.  Although the tools that currently populate the MBONE rely on
loose-style session control, in the past most experimentation has taken
place with tightly controlled session models -- though this is clearly
changing.  The Group speculated that the predominance of tight-control

                                   2





systems may be a function of the interest in supporting ``coordinated''
telecollaborations, which are readily modeled using a tight-control
framework, whereas the emergence of loose-control systems may be a
reflection of the relative ease with which they are implemented.

Systems were clearly differentiated in their approaches to
interconnectivity among participants, both for session and for media
topology.  In certain cases, symmetry exists for N-way communication
capabilities, while in other cases conferees are asymmetrically
interconnected, relying on an initiator, moderator, filter/reflector or
a privileged set of designees to coordinate communication on behalf of
others.  Explicit versus implicit communication is another
distinguishing feature; this relates to whether or not the session has
policies attached to it, such as who dictates membership rules, the
extent to which session information is disseminated or if participant
information is meant to be kept globally coherent.  Finally, it was
observed that the decision to model the system in a centralized or
distributed fashion influenced the degree of messaging synchrony and
causality.

Group Scope, Framework and Functional Taxonomy

There was rough consensus on the definition of conference control as the
management and coordination of multiple sessions and their multiple
users in multiple media.  It was also agreed that the focus of the Group
is to design a ``session layer'' protocol to perform these functions.
However, the Group debated the utility of designing a
``teleconferencing'' session protocol specifically for the coordination
of users' ``media'' versus designing a Group negotiation protocol that
is extensible to act as a conduit for media details.

The Group recognizes that it cannot set out to support all conferencing
scenarios.  However, it proposes to support one loose style protocol (a
la Xerox PARC's nv, INRIA's ivs, BBN's dvc, LBL's vat, UMass' nevot) and
one tight style protocol (for negotiated and potentially private
sessions).  How loose and how tight?  To answer this, the list of
conversation styles must be mapped (from the last IETF Minutes) into
their underlying CONFCTRL session protocols.

As an example of how a tight-control approach to session management
might integrate with already existing MBONE tools, an X-based version of
ISI's MMCC conference control tool was demonstrated at the IETF. MMCC
was used to explicitly invite a specific set of participants (versus
having a wide-open session), to distribute multicast addresses and a
shared encryption key among those participants, and to initiate as well
as tear down sessions comprised of nv, vat and/or BBN's newly released
PictureWindow.

Although it was emphasized that the goal of the Group is to design a
session protocol, the Group conceded that there is a need for a common
framework within which it can talk about conferencing control.  The
framework that arose from discussion, looked as follows:


                                   3






           User A                                       User B

       +-------------+                              +-------------+
       |             |                              |             |
       | Application |                              | Application |
       |             |                              |             |
       +------+------+                              +------+------+
              |                                            |
       +------+------+                              +------+------+
       |             |                              |             |
       |   Session   |<----------------------------->|   Session   |
       |             |      "Session Protocol"       |             |
       +---+--+--+---+                              +---+--+--+---+
          /   |   \                                     /  |   \
         /   ...   \                                   /  ...   \
   +-------+     +-------+                       +-------+    +-------+
   | Media | ... | Media |<---------------------->| Media |    | Media |
   | Agent |     | Agent |     "Media Stream"     | Agent |    | Agent |
   +-------+     +-------+                       +-------+    +-------+



The premise is that the session protocol would be distributed in nature,
and would accommodate multiple user sessions (even though the diagram
depicts only two conferees).  There is a firm separation between the
session protocol and media transport protocols.  Thus, it is immaterial
whether the media transport is packet-based or analog.  Generic session
state would include membership and policy information.
Application-domain specific state might include media interconnectivity
(topology) and media configuration (data encodings, rates).  Although
needing further refinment, the list of session functionality provided to
the end systems and reflected in the session protocol would encompass:


   o Create/Destroy Session
   o Add/Delete Member
   o Set Policy

      -  Who may join
      -  Who may invite
      -  Who may set policies
      -  Etc.

   o Add/Change Application-Domain Specific State

      -  Media interconnectivity
      -  Media configuration

   o Floor Control?
   o Prescheduling?


Polling the interest of the BOF participants, it was found that 75% were

                                   4





interested in solving the session protocol problem, 40% also would be
interested in defining or standardizing the
media-agent-to-session-entity interface, and 30% were interested in
configuration management issues.

Terminology

It became evident that there are no set definitions for terms such as
conference, connection, session, media agents, etc.  Many of the systems
presented during the BOF and described in the templates used these terms
differently.  Thus, a CONFCTRL terminology reference guide needs to be
developed.

The Group had been interchanging the phrases session control, session
management, connection control and connection management, but later
agreed that ``connection'' is too ambiguous since it is used at any
number of levels in the protocol stack.  Connection was replaced by the
term ``session'', and was broadly defined as an association of members
for control purposes.  However, it was later argued that session looks
too much like an OSI term.  The term ``conference'' was also felt to be
too application specific.  Therefore, the Group is open to suggestions
for a better name.

It was suggested (although not entirely resolved) that ``media agents''
handle the media specifics associated with a session.  ``Media'' could
be considered any data streams that involve communication.  It was also
suggested that floor control is deemed the responsibility of a media
agent when it concerns a single media agent, but the responsibility of
the session entity when it requires coordination across different media
agents (e.g., video to follow audio).

The Group also differentiated between two meanings of configuration; the
static end-system description, including hardware and software
capabilities, and the per-session description.

Liaisons

The CONFCTRL Group is committed to tracking the progress of related
efforts, both within and outside the IETF. An important IETF linkage is
to leverage off ongoing work in the Audio/Video Transport Working Group
(AVT), which is nearing completion of the Real-time Transport Protocol
(RTP) specification.  During the first two AVT sessions, there was
considerable discussion about RTCP, the control protocol associated with
RTP. Certain functions in RTCP were felt to violate ``layering''; they
do not belong in the transport, but would live comfortably within the
session level, e.g., text strings of session participants.  The Group
will need to follow closely the outcome of these developments,
especially if certain services are assumed to percolate into the session
layer.

The MBONE is another strategic testing ground for a CONFCTRL solution,
although its use should not preclude use of these ideas elsewhere, nor


                                   5





should these ideas be tailored specifically to the MBONE. By mentioning
MBONE it is really meant that the Group expects, in the long term, to
have access to networks that support multicast and in the longer term to
support real-time services.  The general Internet should suffice for
now.

Individuals who volunteered to track developments in related areas
include:


  Ruth Lang             Directory Services

  Hans Eriksson         Multicast Developments

  Fengmin Gong          Resource Management/QoS

  Steve Casner          Audio/Video Transport

  Eve Schooler          Audio/Video Transport

  Paul Lambert          Security

  Stuart Stubblebine    Security

  Yee-Hsiang Chang      ATM

  Peter Kirstein        MIBs


Action Items


   o Make CONFCTRL bibliography available.
   o Documentation:

      -  Terminology reference guide.
      -  Refinement of functional taxonomy.
      -  Turn Minutes into issues/framework document.
      -  Mapping of conversation styles into session protocols.
      -  Collect suggestions for a Group name change.


Attendees

Lou Berger               [e-mail]
Monroe Bridges           [e-mail]
Al Broscius              [e-mail]
Randy Butler             [e-mail]
Yee-Hsiang Chang         [e-mail]
Brian Coan               [e-mail]
Richard Cogger           [e-mail]
Simon Coppins            [e-mail]
Dave Cullerot            [e-mail]

                                   6





Steve DeJarnett          [e-mail]
Ed Ellesson              [e-mail]
Chip Elliott             [e-mail]
Hans Eriksson            [e-mail]
Deborah Estrin           [e-mail]
Francois Fluckiger       [e-mail]
Jerry Friesen            [e-mail]
Fengmin Gong             [e-mail]
Kenneth Goodwin          [e-mail]
Mark Green               [e-mail]
Russ Hobby               [e-mail]
Don Hoffman              [e-mail]
Frank Hoffmann           [e-mail]
Michael Khalandovsky     [e-mail]
Peter Kirstein           [e-mail]
Jim Knowles              [e-mail]
Lakshman Krishnamurthy   [e-mail]
Giri Kuthethoor          [e-mail]
Paul Lambert             [e-mail]
Ruth Lang                [e-mail]
Patrick Leung            [e-mail]
Allison Mankin           [e-mail]
Donald Merritt           [e-mail]
Paul Milazzo             [e-mail]
Robert Mines             [e-mail]
Joseph Pang              [e-mail]
Geir Pedersen            [e-mail]
John Penners             [e-mail]
Bala Rajagopalan         [e-mail]
Michael Safly            [e-mail]
Eve Schooler             [e-mail]
Michael St.  Johns       [e-mail]
Stuart Stubblebine       [e-mail]
Sally Tarquinio          [e-mail]
Claudio Topolcic         [e-mail]
Mario Vecchi             [e-mail]
Abel Weinrib             [e-mail]
John Wroclawski          [e-mail]
Yow-Wei Yao              [e-mail]



                                   7