Research On The Model Of Overlay Multicast For Streaming Media

Overlay multicast is one application layer data transmission model of multicast which utilizes the function of end-to-end data transmission between users. Overlay multicast is becoming more and more important as large-scale streaming media transmission scheme for its not modifying the infrastructure, easy to be implemented and flexible structure compared to IP multicast's problems on protocol complexity, scalability and deployment/management costs.For overlay multicast on streaming media transmission, the issues of capacity/bandwidth limitation, topology-aware, multicast tree depth, overlay delay and load-balancing should be considered. This paper proposes one layered overlay multicast model according to the key issues of overlay multicast. The overlay network is layered as clusters according to the users'location. This layered cluster structure could make the overlay network topology-aware and improve the efficiency on utilizing network bandwidth. The overheads of network construction and management would be decreased via the layered structure. The method based on network coordination is adopted to implement layered clustering based on nodes'location, and the experiments show this method is valid.The algorithm based on fitness-based parent node search is proposed to construct optimized intra-cluster overlay multicast, and the combined fitness function is defined to implement multiple-object optimization on load-balance and tree depth performances. The theory analysis and experiments show that proper tradeoff could be achieved between the two performances of load-balance and average tree depth via change the combine coefficient which defined in the combined fitness function. The multiple-object optimization could gain better general performance than single-object optimization. The two-step method is proposed to treat with the nodes'capacities limitation and bandwidth limitation respectively to decrease the overhead on bandwidth measurement during parent nodes search.In order to decrease the computing overhead of construction, the expected fitness based preferential random walk (eP-RW) algorithm is proposed for parent nodes search during the construction of intra-cluster overlay network, in which the theory of random walk sampling is use for reference. The simulations and experiments show that the construct algorithm based on eP-RW search could gain very close performance to the algorithm based on global search with the random walk length O(), in which is the average depth of the multicast tree. The eP-RW based algorithm could gain proper tradeoff between the performance optimization and computing overhead. is direct proportional to the logarithm of N (nodes number), and effected by the heterogeneity of nodes'capacities and coefficient of combined fitness. More large the coefficient value and more heterogeneous the nodes'capacities, more small the value, and more short random walk length would be needed to achieve the same performance via eP-RW.One prototype system of overlay multicast is realized in this paper, and the experiments based on prototype system prove the model and construct algorithm of intra-cluster overlay multicast are valid and feasible.