Research On Application Layer Multicast Algorithms

With the rapid development of Internet, multicast communication is required by more and more applications. IP multicast is a native multicast means which implements multicast in the network layer. However, its inherent drawbacks incur the serious deployment limitations of IP multicast. In this situation, application-layer multicast is proposed under holding the promise for realization of large scale Internet multicast services. Application-layer multicast is such a data structure that provides multicast functionality in the overlay networks through unicast connections among group members in the underlying layers. In contrasted to the IP multicast, the multicast functions in application-layer multicast are implemented by end hosts instead of intermediate routers. Therefore the deployment is easy for application-layer multicast. However, the resource and the multicast implementation in the context of overlay networks are different from that in the convention networks that presents new challenges and new problems to solve. How to improve the multicast-related performance of application-layer multicast is important for realization of large scale Internet multicast services.This dissertation focuses on how to improve the performance penalty of application-layer multicast based on two overlay topologies: tree and m-D mesh. Our objectives in this dissertation are the efficient in terms of short multicast delay, scalable in terms of group size, fully distributed and quality of service guaranteed application-layer multicast services.Firstly, this dissertation studied the application-layer multicast protocol based on the tree overlay topology. Focusing the improvement of efficiency and scalability, a Fibonacci series based hierarchical application-layer multicast protocol is proposed in the context of tree topology. A set of novel location-aware algorithms is proposed to construct the hierarchical architecture. All group members are partitioned into different layers and different clusters in each layer. The considering of underlying network properties when clustering reduce the packet delivering on the costly links. Moreover, a Fibonacci multicast tree is constructed by a Fibonacci series based multicast algorithm for cluster members to multicast efficiently. The considering of underlying network properties and the construction of Fibonacci multicast tree enable HFTM better delay performance compared with traditional application-layer multicast protocols.Secondly, the dissertation observed that the end host capacity constraint can impact the multicast efficiency which ignored by many designers. An efficient capacity-aware application-layer multicast algorithm is proposed based on the tree overlay topology--DTALM In DTALM, a degree-constrained Fibonacci tree is constructed in order to achieve better delay performance under considering the capacity constraint of end host. The group members in each local area are firstly constructed into a low-level Fibonacci multicast tree. All roots of low-level Fibonacci multicast tree are then constructed into a high-level Fibonacci multicast tree. The simulation results illustrate that DTALM in the dissertation have better delay performance compared with traditional application-layer multicast protocols.Moreover, the dissertation studied the application-layer multicast protocols based on the m-D mesh overlay topology. A distributed and scalable application-layer multicast protocol is proposed on top of m-D mesh overlay topology-DSALM. All multicast group members are organized into a two-layer hierarchical architecture by utilizing the location-aware algorithms. As to the multicast routing, different routing schemes are utilized on low layer and high layer of the two-layer architecture. The novel multicast architecture and the routing scheme enable DSALM better efficiency and low maintenance overhead. Moreover, the scalability of DSALM is dominant when it is used in the multi-source group applications.At last, for the quality of multicast services, the dissertation studied the issue of balancing the decrease of end-to-end delay and the elimination of delay jitter caused by the fluctuation of network throughput during the packet transmission. Two novel balance schemes: PB-B and PB-R are proposed to achieve the improved balance between the continuity and real-time performance. The two schemes both utilize the instantaneous network situations to make corresponding adjustment. PB-B introduces an adaptive buffer for each receiver and adjusts the buffer volume according the information achieved by probing scheme. PB-R introduces a (σ,ρ) regulator for each receiver and adjusts the value ofσandρsimultaneously. The two schemes both aim to achieve the continuous and real-time playback service. The simulation results proved that the proposed algorithms can achieve improved balance, namely the continuous and real-time playback at the receivers.