Architecture And Mechanisms For Improving Service Capability Of P2P Streaming Systems

At present, live streaming and video-on-demand applications are becoming one ofthe “killer” Internet applications. In streaming systems, the bandwidth consumption atstreaming server increases with the number of customers. Although it is easy to fufillcentralized management in Client/Server (C/S) based streaming systems, the largevolume of bandwidth consumption at streaming server makes this kind of systemsunscalable. In recent years, Peer-to-Peer (P2P) technique based streaming systemshave been proposed and deployed. Peers (customers) self-organized to an overlaynetwork which is independent with the physical network topology, and then thestreaming data is distributed based on the overlay structure. Owing to the highscalability, high efficiency of bandwidth utilization and lower bandwidth consumptionat streaming server, P2P streaming systems have been widely adopted and deployed bymajority of streaming service providers. However, it is hard to claim that the currentP2P streaming schemes are perfect. The characteractics of bandwidth heterogeneousand dynamics of peers make the uncertainity of service ability provided by peers.Especially with the development of mobile Internet techniques since2010, theheterogeneity and diversity of users are becoming increasingly prominent. On the otherhand, the streaming applications pose some strict requirements on network bandwidth,latency of streaming data retrieval, etc. Therefore, P2P based streaming distributiontechniques and systems face a number of challenges, such as load balance, data searchand location efficiency, improvement of playback quality at peers and bandwidthconsumption minimization with satisfying peers’ requirement. This dissertation isdevoted to address these technical challenges which are relevant to the servicecapability of P2P streaming systems by conducting the following research projects.1. Load Balancing based Bandwidth Request Allocation. The heterogeneity ofbuffered streaming data and playback position at peers may easily lead to loadunbalancing problem. If this problem occurs frequently in the overlay network, largevolumn of streaming data distribution could be restricted, and thus the Quality ofService (QoS) at peers is deteriorated. This dissertation analyzes multiple peers request streaming data from their neighboring peers based on game theory. The optimalbandwidth request allocation policy is obtained through searching the NashEquilibrium of the game. In addition, a deployable bandwidth request allocation isproposed based on this equilibrium solution. Experimental results show that theproposed bandwidth request allocation policy can efficiently avoid the occurance ofoverloaded peers in overlay networks and decrease the latency of streaming dataretrieval at peers. The service ability of P2P streaming system can be effectivelyimpoved.2. Efficient Streaming Data Location for P2P Live Streaming System. Since thestreaming applications are latency sensitivity, it is important to improve the streamingdata search and location effeciency at peers. This dissertation proposes a one-hopbased Distribtued Hash Table (DHT) structure, called StreamDHT. In order toguarantee that all peers in StreamDHT structure can be accurately and efficientlyinformed of joining/leaving event messages, StreamDHT adoptes an enhanced eventdistribution scheme-EDRA+. Moreover, in order to guarantee the success ratio ofstreaming data location and control the traffic consumption on Indexes maintenance,this dissertation proposes a new distributed Index mapping and managementmechanism. Theoretical analysis and simulation validation show that StreamDHTstructure can significantly improve the streaming data location with lower maintenancetraffic overhead.3. Improving Playback Quality in P2P Live Streaming System. Due to thebandwidth resource heterogeneity and dynamic characteristics at peers, it ischallenging to provide satisfactory playback quality to large number of peers. Thisdissertation proposes a new P2P live streaming architecture based on mesh and DHThybrid overlay network. According to the current data re-location requirement andpeers’ online duration, a portion of peers which can provide stable streaming serviceare selected to form the StreamDHT structure. The rest of peers self-organize to form amesh based overlay network. Based on this kind of hybrid overlay network, thisdissertation proposes a joint data scheduling and DHT data re-location policy. In orderto avoid playback quality deterioration which is caused by that the needed framescould not be obtained before the playback deadline, peers can request the neededstreaming data through using different data retrieval methods based on the emergency and decoding importance of frames. Theoretical analysis and experimental resultsshow that the proposed live streaming distribution architecture can effectively improvethe playback quality at peers with trival extra overhead.4. Optimal Neighboring Peer Selection and Bandwidth Allocation in Layered P2PVideo-on-Demand (VoD) Systems. Through combining layered video coding with P2Ptechniques, Layered P2P VoD systems can provide differentiated video quality to peerswith heterogeneous bandwidth resources. However, the time of entering the system andcurrent playback position are heterogeneous at different peers. Thus unreasonableneighboring peer selection and bandwidth allocation between peers can easily lead tosevere bandwidth consumption at streaming server. This dissertation formulates anoptimization model of bandwidth consumption on streaming server. The theoreticalminimal bandwidth consumption can be obtained by finding the optimal results of thismodel. In addition, online duration similarity based neighboring selection and thesingle/cross video layer banwidth allocation polices are proposed. These proposedpolices are easy to be deployed in a practical systems. Experimental results show thatthe proposed polices can efficiently decrease the bandwidth consumption, and theoperation cost at streaming service provider could be decreased.