Regular Structured Multicast Service Based On P2P Clustering

In recent years, the growth in demand for streaming media application of Internet users is rapidly and the increasing resources of the streaming media service are difficult to meet the demand for streaming media in both scale and quality. That is why P2P streaming multicast, as one of the solutions of the problem, is gaining more and more popularity. P2P multicast can significantly reduce dependency of streaming multicast on the service resources by constructing the application layer overlay network which uses the node capacity of service supporting by the transmissions between nodes. Existing P2P multicast streaming media business platform constantly improves technology to support for large-scale, high-quality streaming media multicast services, and achieves remarkable results.However, the robustness and scalability of streaming media multicast system is still the problem not completely solved by current multicast system and the theory. The robustness mainly concerns that the frequently joining and leaving of nodes make an impact on the multicast quality. And scalability mainly concerns that the increasing of the scale of nodes makes an impact on the multicast quality. Another problem is cross-network transmission wildly existing in current P2P multicast systems, which leads to transmission losses of peers. Current approaches can not sovle these problems satisfactorily.This paper proposes Regular-Structured Multicast (RSM) model, including three theoretical parts:overlay network model of RSM, P2P clustering service and stream service. A P2P streaming system based on RSM model, namely ClusterStream, is also introduced in this paper.Overlay network model is the core component of RSM model. It is a multiple-trees based overlay network model, and defines the static topological structure of RSM. The most significant difference between overlay network model of RSM and existing multi-tree models is that RSM does not generate forwarding trees directly, but indirectly by maintaining Regular-Structured clusters first. Several technical problems are overcome. First, Regular-Structured cluster is hierarchical structured, and forwarding trees of the whole overlay network can be smoothly and automatically reconstructed by structure transformation of clusters. This feature can reduce the complexity and the overhead of overlay network management. Second, Regular-Structured clusters force peers and clusters persisting constant number of download and upload transmission stream-links. Simulations show that P2P multicast systems that keeping peers uploading load balanced suffers less risk of getting media playback interrupted, and better robustness of service. Further, to dealing with heterogeneity of peers'service ability, two extension schemes are also proposed, helping overlay network of RSM taking more efficient use of peers'ability and offering media stream service of different service level for peers with different service abilities.P2P clustering service provides topology-aware capability to RSM. First, we analyze the phenomenon of cross-network P2P transmission, as well as the possible bandwidth loss it cause. Then we prove the rationality of mapping P2P distance information to a metric space in both theoretical and experimental ways, and show using clustering space is more accurate and requires less calculation cost than current approches using coordinate space by comparative simulations. Finally, P2P clustering service for RSM is presented. It is designed as a service that manages clustering view of peers by clustering initiating and optimizing operations, and offers public interfaces for clustering view querying. A clustering view covers all peers from different P2P overlay networks and applications, so P2P clustering service provides an integrated topology-aware surport for multiple multicast channels of in a RSM system.Stream service is a high performance service that maintains clusters in RSM system dynamically. Based on the global stream algorism, advanced algorisms in stream service can make clusters to expend and contract with lest stream operations when peers join and leave the multicast channel. To further reduce overheads of centric servers in stream service, asychonous stream operations are designed, which peers execurate without sychronizing middle status with servers. Accelerration mechanism for stream initiation is also adopted to reduce the risk of media playback interruption.Finally, we implement the ClusterStream service platform based on the Regular-Structured multicast model. The platform uses WCF, ADO Entity framework and can be implemented on IIS6 and SQL Server. In addition, ClusterStream platform is able to monitor and manage in Web-background, with high ease of use. Our performance experiments show that the platform has the ability to support multicast for large-scale Internet services.