Sigcomm'95 Workshop on Middleware: Position Paper

FLOOR CONTROL FOR NETWORKED MULTIMEDIA APPLICATIONS

Hans-Peter Dommel and J.J. Garcia-Luna-Aceves

{peter | jj}@cse.ucsc.edu

Baskin Center for Computer Engineering & Information Sciences
Computer Communication Research Group (CCRG)
Applied Sciences, University of California Santa Cruz, CA 95064

POSITION PAPER for the SIGCOMM'95 Technical Symposium (Workshop on Middleware)

KEYWORDS: Concurrency control for shared multimedia objects (floor control), Computer-Supported Cooperative Work (CSCW) with multimedia support.

INTRODUCTION
A migration of multimedia platforms from stand-alone and single-user scenarios to heterogeneous networked environments can be observed. One of the major reasons for networking multimedia applications is the opportunity to facilitate CSCW with support of multiple media, enhancing the scope of applications from a local platform to a local-area or wide-area collaborative setting in dynamically rearranging groups. For such applications, new approaches to manage the aggregation of sessions and coteries are needed, as well as protocols to handle the variety of multimedia streams serving users in information exchange and collaborative work.
Conflicts in accessing shared resources may arise, when users from different sites want to use the same resource. Such conflicts need to be mitigated or ideally avoided a priori. That may be trivial for smaller sessions with only a few participants, where ``tight control'' on group activities is established by more personal interaction. However, in larger and more anonymous collaboration scenarios, such control of who uses and manipulates which resource when, can lead to confusion and inefficiency.
Enhancing the scope of multimedia applications into a distributed setting creates hence many new problems with respect to concurrency control, coherency control, access control, scheduling and allocation of resources, synchronization and replication of data. While first attempts have been made to characterize access control in distributed multimedia settings [14], concurrency control and other issues have been barely examined in the context of collaboration and dynamic fine-grained data sharing.
Floor control is a novel concept for coordination of groupware [5] activities in networked multimedia applications. The main idea is to give "turn-taking" structure to online meetings whose coordination lies otherwise on mixed and often misinterpreted signals in order to resolve the problem who is authorized to work on what data at which point in time. Until now most information sharing is only possible on static results of individual work efforts, and floor control puts such endeavor into the much broader context of joint and dynamic work.
Important roots of floor control lie in psycholinguistic research on turn-taking behavior in discourse [8, 9]. However, since the CoLab system [15], one of the fundamental studies on CSCW with "manually" negotiated floor control via signals warning about access conflicts in groupwork, not much progress has been made for automatic [12] and flexible floor control in the context of computer-mediated communication [8].
A floor designates traditionally the right to speak in an assembly and is granted by a conference chair. For multimedia-supported CSCW the chair can be either a designated participant in an online session or a system process, called the floor agent. The semantics of a floor is enhanced to the right to speak or take action on shared data in a distributed software environment. Managing floors bears similarities to the problems of managing mutually exclusive access to shared zones or objects, cf. ground control in traffic studies or semaphore-regulated access to shared memory in multiprocessor systems. However, its most prominent technical heritage is concurrency control methodology known from transactional database access. Depending on its realization, floor control is located anywhere between the transportation and application layer, yet typically existing applications integrate floor control in the application layer.
LIMITS OF EXISTING SERVICES
Floor control is currently featured only in few collaborative multimedia applications [7]. Floor control, as component of a middleware infrastructure, interfaces with session control [4], i.e. management of online meetings from setup, dynamic joining and leaving, to termination. Such sessions can vary in scale and configuration depending on many parameters characterizing participants and media in use. Applications featuring floor control do not scale with the number of "collaboratees" and are often restricting sharing to simple objects like tele-pointers. There is not much adjustability or flexibility with respect to fine-grained tool-sharing and usage policies, hence limiting the interactional range among conferring users.
PARADIGMS FOR FLOOR CONTROL
Floor control is a part of middleware infrastructure interfacing with session control. Even though it is an application-level concept, it can be applied as well to channel access in end-to-end services. Several paradigms for supplying a notion of floor control can be identified:

tapping the X-server and filtering events handling specific shared objects, granting control to specific users outside of applications [1],
integrating floor control into a "collaboration-aware" application with session control as an integral part of the application per se [10],
interfacing multimedia objects and resources within applications to outside agents [3], which relay control information across distributed sites and negotiate schedules for resource access.

SYSTEM REQUIREMENTS FOR FLOOR CONTROL
Ideally approximating a face-to-face encounter, floor control needs to provide responsiveness to user's requests for certain floors. However, an adaptive and comprehensive control service must obey many causally related parameters of the session context, causing protocol overhead. Not only "psychological" user constraints, but system Quality-of-Service (QoS) requirements need to be observed as well. Media can be lossy or lossless and impose real-time requirements or allow for jitter. Floor control strategies need to adjust to such lower level service guarantees and characteristics.
The following requirements need to be observed in realization and provision of floor control [2]: it must scale with the number of participants, adapt to session and resource specifics, establish a balanced and fair procedure to grant control rights over resources to individuals across the session community, and allow for extensibility to new tools. Furthermore, it must work distributedly, yielding resiliency, efficiency and hence responsiveness. The protocol serving behind the implementation must ensure correctness, liveness, and preemption in floor assignment. Ultimately servicing users, usability of automatic floor control provision must be ensured to yield acceptance. This can be achieved by making floor control adjustable and intelligible through the user-interface, yet its operation transparent by hiding internal control actions. Of course, such service can only work within the framework of a standardized collaboration architecture, allowing for interoperability between shared applications of different kinds across heterogeneous platforms. However, floor control must also scale with the "richness" of the collaborative environment and allow for self-contained control, when for example only audio-channel control is available.
Session control, with floor control as its counterpart, must observe similar requirements [13]. A typical phenomenon of face-to-face meetings is that people in larger aggregations tend to establish side-conversations without wanting to drop out from the larger context. Such recursive and temporary group-dynamics has to be reflected in hierarchical conferencing models [11] and adjunct floor management.
DIFFERENCES TO TRADITIONAL CONCURRENCY AND ACCESS CONTROL
Similar to concurrency control (CC) used in (distributed) database systems, multiple requests for interleaved access of a shared resource create a critical region in floor control (FC), e.g., when several users try to attain the speaker floor of an audio-channel and intermixing of multiple audio streams on the same channel occurs due to transfer delays and uncontrolled channel usage, making conversation impossible. Like a transaction, an action on specific media is started with floor-attribution and completed with floor-release, and must be atomic, complete and unique with respect to interferences from other sites to ensure smooth workflow. Yet there are major differences between FC and CC:

FC must organize a more dynamic session context with flexible strategies, whereas CC deals with rather predictable events and resources.
FC mediates in interactions among users, whereas CC regulates transactions between users targeting the same database.
FC must allow for multiple access into a critical region for certain tools (e.g. tele-pointers), and objects can have finer granularity and access distinctions, in opposition to CC with unary mutual exclusion and simpler data-structures.
Recoveries and rollbacks are hardly possible for most multimedia interactivities due to real-time constraints and data volume. Multimedia collaboration is often audible and visible and hence more tangible than transactions.
Group-dynamics between collaborating people causes strong variance on the intensity of contention for data-access, requiring adjustable policies for buffering and scheduling of shared events.
CC works generally optimistic, i.e., a transaction monitor identifies conflicts at commit time. FC must operate more pessimistic due to the interactive nature of most media.

APPROACHES TO PROVIDE FLOOR CONTROL
Provision of FC needs to observe the outlined requirements from the system's as well as the user's perspective. Since FC conceptually integrates previous methodologies from database and operating systems technology, placing them into a new context with more complex and stringent requirements, a thorough framework on FC promises a viable point of departure for "Multimedia Collaboration Systems". New paradigms of user-interfaces overcoming the limitations of video conferencing [6] and allowing for telepresence coming close to the quality of face-to-face meetings need to be renditioned. A first step towards providing more elaborate FC is the realization of distributed agents tracking a set of sharable resources by registering, filtering and locking them based on session-wide events and actions. Such "floor agents" interface and confer with the session control agent on each site, which maintains a directory of participants on each level of a session hierarchy, controlling group membership and session characteristics. For the sake of resiliency, drop-outs on floor-carrying channels or temporary and complete withdrawals of specific sites are not to affect the remaining session, i.e., "orphaned" floors and truncated sessions need to be embedded in fault-tolerant session management.
CONCLUSION
Much research is necessary not only to find the "golden cut" between the application level as well as network and operating systems service levels, but also to standardize a concept like floor control across the multitude of platforms and software environments. Floor control can aid in supplying multimedia-CSCW interoperability, fostering "collaborability" among applications. As a paradigm to secure or publicize data across networks, it enhances authenticated data-access. However, with every opened door to a common data-pool, new network security holes emerge and further work is needed with respect to encryption, authentication and auditing in collaborative scenarios. Middleware like floor control, intended to support a larger spectrum of collaborative works, will be quintessential when wireless devices like Personal Digital Assistants allow for a broader range of interactional usage of shared data-banks.
REFERENCES
[1] E. Craighill, R. Lang, M. Fong, and K. Skinner. CECED: A system for informal multimedia collaboration. In Proc. of the ACM 1993 Multimedia Conference, Anaheim, CA, August 1993.
[2] H.-P. Dommel and J.J. Garcia-Luna-Aceves. Design Issues for Floor Control Protocols. Proc. Multimedia and Networking '95, IS&T SPIE, San Jose, CA, February 1995.
[3] E.A. Edmonds, L. Candy, R. Jones, and B. Soufi. Support for collaborative design: Agents and emergence. CACM, 37(7):41--47, July 1994.
[4] W.K. Edwards. Session management for collaborative applications. Graphics, Visualization & Usability Center, College of Computing, Georgiatech, Atlanta, Internet Draft, 1994.
[5] C.A. Ellis, S.J. Gibbs, and G.L. Rein. Groupware -- some issues and experiences. CACM, 34(1):38--58, January 1991.
[6] E.A. Isaacs and J.C. Tang. Why do users like video? Studies of multimedia-supported collaboration. Computer Supported Cooperative Work (CSCW), 1(3):163--196, 1993.
[7] P.S. Malm. The unOfficial Yellow Pages of CSCW -- Groupware, prototypes and projects. Technical report, University of Tromso, Norway, January 1994. URL http://www.tft.tele.no/cscw/.
[8] A. McKinlay, R. Procter, O. Masting, and R. Woodburn et. al. Studies of turn-taking in computer-mediated communications. Interacting with Computers, 6(2):151--171, June 1994.
[9] D.C. O'Connell, S. Kowal, and E. Kaltenbacher. Turn-taking: A critical analysis of the research tradition. Journal of Psycholinguistic Research, 19(6):345--373, Nov. 1990.
[10] A. Pang, C. Wittenbrink, and T. Goodman. CSpray: A collaborative scientific visualization application. In Proc. Multimedia and Networking '95, San Jose, CA, February 1995. IS&T SPIE.
[11] P.V. Rangan and H.M. Vin. Multimedia collaboration as a universal paradigm for collaboration. In Multimedia - Principles, Systems and Applications, pages 3--15. Springer-Verlag, April 1991.
[12] S. Sarin and I. Greif. Computer based real-time conferencing systems. In Computer Supported Cooperative Work: A Book of Readings, pages 397--420. Morgan-Kaufman, 1988.
[13] E. M. Schooler. The impact of scaling on a multimedia connection architecture. ACM Multimedia Systems, 1:2--9, 1993.
[14] H. Shen and P. Dewan. Access control for collaborative environments. In Proc. CSCW'92, pages 51--58, November 1992.
[15] M. A. Stefik and G. Foster et. al. Beyond the chalkboard: Computer support for collaboration and problem solving in meetings. In Computer-Supported Cooperative Work: A Book of Readings, pages 335--366. Morgan-Kaufman, 1988.