Application of media channels to large-scale group communication
While computer networks have been of low capacity or poor reliability, their use for real-time communications for human interaction was considered infeasible on any large scale, since audio and video data consumed vast amounts of memory and processing capacity. However, recent and forthcoming enhancements to network quality, along with advanced compression algorithms, falling costs of memory and fast processors, have allowed such applications to develop, as well as protocols to support them. On top of this, global interconnectivity of local networks has improved, giving scope to expand human communications applications to wide areas, and with many participants. Such group communication applications have various networking Quality of Service (QoS) demands, such as lower bandwidth but high reliability. The mahor challenge in multimedia group communications is therefore about developing network services to support group communications with varying QoS, in order to allow groups of people, potentially very widely distributed around the world, to interact through computers of varying abilities, using combinations of media such as video, audio, and shared editing of documents. The Media Channel (MC) model is an attempt to unify the control and management of applications of different media types. It provides common group management facilities for all types of application, and allows participation in an application to be expressed simply as contribution (using a source) and/or observation (using a sink). This further allows a multimedia group application to be built up from several ‘unimedia’ group applications (the ‘channels’) through the use of a user agent which manages and simplifies a user’s participation in several applications. Participation in multimedia group applications (or ‘sessions’) can then be configured automatically for several users by describing the channels of the session to each involved user agent. The description itself can be supplied through a channel, with the agents acting as sinks. The user can control participation in all the channels of the session by (de)activating the agent’s session sink, and while joined, session administrators can inform all session participants of new channels to which they are invited. The MC model's main shortfall is its use of a central manager — an architecture which does not scale well to many participants, and which may lead to uneven management for widely distributed participants, particularly as group populations migrate. This thesis considers the issues of multimedia group communication, and attempts to realise the MC model. Some earlier definitions of the MC model are refined in the present thesis, and alternative architectures are designed, implemented and evaluated. These new architectures allow the management components of a media channel to be placed near to the participants, and then to migrate as participants leave the components’ areas, and as new ones join in others. At the same time, management information to evaluate policies is compressed and distributed among components. The model is also extended to allow sources to describe their data arbitrarily, enabling sinks to accept or reject them individually depending on availability of local resources.