Research On Streaming Media Scheduling Based On Novel Distributed Video-on-Demand Architecture

Along with the rapid development of broadband network and video coding technologies, streaming media traffics such as Video-on-Demand (VoD) and IPTV have attracted much attention, but the performance of current access network can not meet the controllable, manageable and operable needs of these applications. To face the challenge,―High Performance Broadband Information Network (3Tnet)‖of the National High-Tech Research and Development Program of China (863 Program) launches the project of―system performance and key technologies research on the access and convergence router (ACR)‖. ACR guarantees the user network access bandwidth, provides a novel access network solution architecture supporting broadband interactive streaming media services and is the key equipment in the―China Interactive New Media Network and Traffic Project‖of National Key Technology R&D Program.Streaming media consumes huge bandwidth and requires real-time transfer with a long period of time. The streaming media server is the main factor affecting the quality of service of these applications, which is also true in ACR environment. Streaming media scheduling is responsible for allocating, managing and using the server and network bandwidth, and is the important factor affecting server performance. One of key technologies for ACR research is to design simple but efficient and scalable streaming media scheduling techniques. Coupling with requirements of ACR project, the dissertation proposes a novel distributed VoD architecture and mainly focuses on the streaming media scheduling. To be concrete, the main work and contributions are as follows.A novel distributed VoD architecture (DVoD) supporting large-scale users is devised to overcome the poor scalability of the traditional unicast based VoD system. Based on the idea of edge media server streaming, DVoD uses unicast communication in the backbone network and multicast communication in the access network. Video servers in the access network answer for streaming the media to clients using multicast based scheduling technique. Scheduling is the crux of the scalability of DVoD. Simulation results show that proposed scheduling algorithms can support client heterogeneity and interactivity and reduce the required server bandwidth from O(λ) to O ? ? ? whereλis the client request arrival rate.A scheduling algorithm called Maximum Urgency First (MUF) is presented. The urgency notation is introduced by formulating optimization problem for average access delay. The definition of urgency considers the client number, client waiting time and the required stream length of the video queue. Whenever a channel becomes available, MUF schedules the queue with the largest urgency. According to the different evaluation methods, three implementations are proposed for Batching: MUF-P, MUF-M, MUF-I and three schemes are proposed for stream merging: IMUF-P, IMUF-M, and IMUF-I. Extensive simulation results show that MUF can yield excellent performance in terms of defection rates, average access latency and fairness. A scheduling scheme called Enhanced Patching (EPatching) is proposed to support client heterogeneity. EPatching supports heterogeneous clients by using catching-up streams at different transmission rates and achieves resources sharing ultimately through stream merging. The optimal multicast interval, the average server bandwidth and server bandwidth distribution for EPatching are all analyzed. The cost based implementation called EPacthing-C is also presented and IMUF-I is used to schedule the waiting requests. The efficiency and scalability of EPatching are verifed through extesinve simulation and comparison with other schemes.The client-class based admission control mechanism is analyzed. Assume that there are several client classes. Each class occupies different server bandwidth and is enforced additional threshold. The recursive method calculating the blocking rate is provided to decrease the calculation complex, and experimental results are identical with the analytical model. From the view of maximizing system profits, the revenue model is investigated to derive the optimal thresholds for different client classes. The analytical results show that this scheme performs better than the fully shared admission policy in terms of the system pay-off through insuring the quality of service (QoS) of higher priority clients.A scheduling policy called Regular Multicast Fixed Scheduling (RMFS) is advanced to support the client interactivity. To address the problem of performance degeneration in the context of user VCR operations, RMFS schedules the regular multicast every fixed time interval to insure the existence of the target multicast during the stream merging, and adopts the class-based admission control to guarantee the QoS of interactive requests. Simulation results verify the correctness of the analytical model and the feasibility of the proposed scheme. The extension for RMFS in heterogenous environment is also discussed.