Multiparty videoconferencing is one of the most important applications in the broadband network era. Unfortunately, it is difficult to perform high quality multiparty videoconferencing with computers over the Internet due to its huge demand on system resources and its stringent requirement on timely delivery of media streams. In order to maximize throughput and minimize delay and delay jitters, videoconferencing applications need to adapt to resource fluctuations, heterogeneous characteristics and capabilities of the conference computers, and the dynamically changing participants of the conference....[ Read more ]

Multiparty videoconferencing is one of the most important applications in the broadband network era. Unfortunately, it is difficult to perform high quality multiparty videoconferencing with computers over the Internet due to its huge demand on system resources and its stringent requirement on timely delivery of media streams. In order to maximize throughput and minimize delay and delay jitters, videoconferencing applications need to adapt to resource fluctuations, heterogeneous characteristics and capabilities of the conference computers, and the dynamically changing participants of the conference.

In this thesis, we propose a resource allocation scheme that distributes workload dynamically according to the resource availability at the conference machines. In this scheme, the conferences are managed by the group server. The group server regularly receives, records and analyses the quality of service reports from the conference computers. It calculates resource allocation according to resource availability, relative priorities among the media streams, and minimum acceptable quality of media streams. Resource allocation instructions are then sent to the conference computers. Each computer then adjusts the media streams using media-specific media scaling algorithms according to the QoS directions from the group server.

A Scalable Multiparty Videoconferencing application (SMVS) is implemented as an evaluation prototype on Solaris workstations connected by an ATM network. The empirical results show that our prototype is adaptive to resource fluctuations, and scalable to participating host machines with different capabilities that may dynamic join and leave the conference.