Research On Data Scheduling Strategy For Data-driven P2P Live Streaming

Peer-to-peer (P2P) media streaming system could significantly reduce the consumption of the network bandwidth and the congestion of the local network, so have good scalability and performance. P2P program apply tree structure in the early, not well adapted to the dynamic and the heterogeneity of the network. Then unstructured overlay network was used in P2P media streaming system, having more control message and larger transmission delay. Recently layered composite structure has been widely used in media streaming system. LStream system is exploited by Henan Provincial Key Lab on Information Networking Zhengzhou University,We research the data scheduling algorithm based in this system. In order to make the system meeting the real-time requirements, transmit data efficiently in the neighbor nodes, we do the research.First of all gives the definition and calculation method of data segment priority and peer node capacity, and then propose Least Rarest First Scheduling Algorithm combined with the existing policy Rarest First scheduling algorithm. The algorithm consists of two parts:data scheduling and node scheduling, data scheduling determines the requested data and the order, then select the neighbor node to request data by node scheduling. The importance of streaming data is different, data held only by a few of the neighbors and on the verge of playing is important, that is scarce and urgent data is more important than ordinary data. According to data scarcity and urgency, we determine the priority of the data, and schedule data in the accordance of priority with the descending. The data with the highest priority will be request firstly, so as to improve the overall download rate of the system. After determining requested data, we will find that there are several nodes holding the data simultaneously, this paper consider two key point in node scheduling which is load on transferring data and node capacity. Accept request number is defined to calculate the load on transferring data, the node with minimum accept request number has smaller load on transferring data, so will be chose to request first. Node capacity which mainly affected by node available bandwidth, network link quality and the impact of the request number of nodes is a dynamic change value, using the traditional method of real-time measurement will consume a great deal of resources. This paper estimates capability of peer node real time based on transmission history of the peer, choose the strong peer to request. It can balance the lode of peer, also make full use of peer resources, and improve the overall performance of system by this means. Our performance study shows that under the same conditions, our new technique can dynamically balance the system workload and significantly improve the overall system performance compared with the Random, Round-Robin and the LRF algorithm. The algorithm proposed in paper with higher efficiency, better stability and scalability.