Research On Some Issues Of P2P Streaming Media Sharing System

P2P (Peer-to-Peer) streaming media integrates P2P and streaming media techniques, which makes it possible for a large number of clients to share streaming media based on the current networks. Although P2P streaming media sharing is successful in some fields, it is still with many problems such as scalability, reliability, fault tolerance, QoS, etc. Nowadays, the studies on P2P streaming media sharing include network structure, resource searching, media coding, network coding, buffer management, application layer multicast and so forth. In this dissertation, we focus on some issues of P2P streaming media sharing systems. The main contributions and original ideas included in the dissertation are summarized as follows.Architecture of layered P2P steaming media sharing system. A plain, layered and modularized structure makes it easy to add new modules or delete old ones, and shows good scalability. This dissertation introduces a layered P2P steaming media system, which is composed of P2P fundamental layer and P2P application layer. The system includes three modules: P2P fundamental network, search tree, and mesh-based multicast architecture. This dissertation describes the function of each layer and module, as well as the relationship among them.Unstructured P2P message spread mechanism based on prediction. The message spread is the basis for finding peer nodes, disseminating information, searching information in the P2P network. The existing message spread methods are mainly based on blind flooding and random walk. To guarantee high message cover range, these methods have to increase the amount of redundancy messages and waiting time. This dissertation provides a message spread mechanism based on prediction. This mechanism predicts the validity of the message according to information from neighbor nodes. If the message waiting for sending is invalid, it will not be send. The advantage of this mechanism is reducing invalid messages in the system while guaranteeing high message covering and low waiting delay.P2P search algorithm based on weighted search tree. In the unstructured P2P system without the central node, because of high heterogeneous characteristic of the network, high amounts of nodes with dynamic characteristic, and high complexity of contents in the system, it is hard to search specific contents in the system. Most of existing search methods is based on blind flooding or gossip. The shortcomings include large delay, invalid messages, inefficient searches and so on. This dissertation provides a search algorithm based on weighted search tree. In this algorithm, a logical weighted k-tree is set up according to the historical hit ratio. The nodes of the k-tree are arranged by the hit ratio, which means the nodes with large hit ratio and long online time are set in the top level of the k-tree. Thus it is easy to determine the message dissemination direction in searching. Moreover, some methods including hot peers caching, remote peer connections, source peer index, and result index are employed to increase search efficiency and to balance the node loads. The proposed algorithm can avoid a large amount of unnecessary messages, which results in high search efficiency and low maintain costs of the search tree.P2P multicast for media on demand based on reliable degree of peers. The application layer multicast is an efficient method for streaming share. Aimed at the reliability and stability of P2P on demand streaming, this dissertation provides a P2P multicast architecture for media on demand based on node reliability and QoS. The reliable degree of each node is calculated based on the failure probability of its parents, and the QoS level of its parents is also assessed in the video on demand service. These two factors are the basis of the parent selection, the allocation strategy of the requested media data, the fault tolerance and the error resilience.P2P streaming media buffer strategy based on user character. In the P2P steaming media sharing system, each peer node caches a limited number of media segments around its current play offset, and provides streaming media service for later coming peers, so sharing streaming media data among peer nodes in real-time is realized. Existing buffer strategies are mainly based on sliding window and data prefetch. However, the media on demand streaming hardly utilizes the buffer data when users perform VCR (Video Cassette Recorder) operations frequently. This dissertation provides a buffer strategy based on the user character, which can increase the hit ratio through buffering hot fragments by peer nodes.Complex network model based on group growing. The P2P streaming media sharing system is an overlay network constructed on the current Internet. The overlay network and Internet are all complex networks. The quality of the network model reflecting the characteristic of real-world network affects the design and validation of the system. This dissertation proposes a complex network model based on group growing, which reflects the static and dynamic characteristics of the network in real-world.Simulation platform for P2P streaming media sharing system. The main problems of existing simulation platforms include difficulty to use, lack of documents and updating slowly, difficulties in replacing underlying networks and other modules, etc. For simulating the P2P streaming media sharing system, the WNSim simulation platform is implemented to simulate different P2P network structures, multicast algorithms of media on demand, search algorithms and underlying networks. The modules in the platform can be added, removed, or replaced conveniently.