Hierarchical Scalable Reliable Multicast Distributed Multicast Voice Control

In recent years,the multicast-based conference that includes a large number of hosts and covers a large scope of network is the main development direction of the multimedia conferencing. However,IP multicast is generally unreliable,disordered and with redundant copies,and there is no reliable multicast protocol designed for multicast-based conference control. As a result,it is quite difficult to realize conference control basing multicast. As to the existing multicast-based conference systems,two choices are widely adopted:to keep the traditional centralized conference control structure with multicast only used for sending media data to those receivers who can't interact with others;or to give up conference control at all with multicast used for interactive loosely coupled sessions including a small quantity of hosts. Neither of the two choices can take the advantages of multicast to improve the scalability and robusticity of multimedia conferencing actually. In order to solve this problem,a reliable multicast communication model for conference control is designed specially. We firstly study the requirements of multicast-based conference control,then analysis the main factors to the performance of reliable multicast. Subsequently we present a new reliable multicast protocol named HSRM (Hierarchical Scalable and Reliable Multicast) according to the receivers' packet loss probabilities. Furthermore,we present a distributed self-adaptive mechanism to construct and maintain the hierarchical structure of HSRM dynamically according to a series of criterions such as distributed,dynamical,fair and so on. At last,the comparisons to several well-known multicast protocols SRM,RMTP and TMTP are given. HSRM constructs hierarchical structure basing multicast communication completely and its structure has no relationship with the senders. The spread of feedback and recovery packets are restricted efficiently. The heterogeneity and the size of multicast sub-groups are also decreased. In a word,the performance of the multipoint-to-multipoint reliable multicast is improved,without bad effect on the robusticity,scalability and flexibility. We then present a HSRM-based distributed floor control model to accommodate the application background of WebLearning+,an IP-multicast based collaborative learning system. This floor control model is flexible,scalable and robust. It supports late joining,early leaving and crash recovery.