Multicast multimedia traffic over internet with quality of service

With the growth of Internet, more and more applications require the multimedia traffic to be delivered to multiple destinations with ensured quality, such as video distribution, audio conference, online lecture and so on. For this purpose, QoS needs to be deployed to provide better service to selected traffic to satisfy the requirements of different applications. Furthermore multicast is needed to provide efficient delivery of network contents to multiple destinations. To ensure the QoS to multimedia traffic, we design a new traffic model, S-BIND, and a new admission control algorithm, GammaH-BIND, to admit online traffic only if their minimum QoS requirements can be satisfied. The new traffic model doesn't require scanning of existing traffic trace. Therefore it can be used on on-line traffic where no traffic trace is pre-recorded. Through simulations, we show that S-BIND traffic model catches the burstiness property of on-line traffic. For admitted traffic, we design a new token bucket based statistical regulator to monitor and regulate the real input traffic. The regulator is composed of the cascade of multiple state-dependent token buckets and a statistical packet handling scheme. Simulation results show that the output traffic of the proposed regulator has S-BIND parameters under the constraint of the desired S-BIND parameters. The regulated traffic won't affect other existing traffic's QoS and can receive its reserved service. For the admitted multicast sessions, we use max-min fair bandwidth allocation algorithm to determine the actual QoS received by destinations of sessions, which ensure the bandwidth of links shared fairly among different multicast sessions. Due to the heterogeneous property of the Internet, the actual QoS received by destinations could be different among the same session and could be better than the minimum QoS requirements used for admission control. We propose a new scalable fair bandwidth allocation algorithm to achieve max-min fairness in the DiffServ framework. The new algorithm has no per-session state information maintained in core routers and distributes the computation among the Bandwidth Broker (BB), edge routers and core routers to improve scalability. The correctness of the new algorithm is proved and its complexity is analyzed.