On Optimization For Large-scale Adaptive Video Multicast In Heterogeneous Environments

With the popularization of access to broadband networks, the demand for network video services is on the great increase in recent years. Video multicast, the particular network video service apt at multipoint data delivery, has been widely applied to distance learning, distributed video conference, digital video surveillance and real-time video distribution. In accordance with the enlarging of video multicast application scale, the capability of video multicast networks and receivers varies greatly. Video multicast has suffered criticism on its performance, such as the complexity of management, the difficulty of Quality of Service (QoS) guarantee, etc. In order to solve these problems, the academic and industrial communities have concerned on adaptive video multicast (AVM) for years, but there are still many theoretical and practical problems left to be resolved, especially for large-scale video multicast applications. In this thesis, we study the optimization for large-scale AVM to make it apply in heterogeneous environments.In this thesis, we first classify and analyze several typical AVM schemes. We find most of these schemes take small-scale video multicast applications as the backgrounds, and they only improve the local performance metrics. To extend the results, a total system utility model for AVM systems is presented in this thesis. And an extendable, hierarchically distributed solution is also proposed, by which we can figure out optimally the bandwidth, the layer number, and the layer rates for each session. The proposed method can handle the problems of feedback implosion, single-node failure, and bad real-time performance caused by centralized adaptive control and decision in large-scale AVM applications. Experimental results and analysis of computational complexity prove the proposed distributed method to be of low complexity, easy in implementation, and well performed in real time.It is known that end-point driven and static configuration agent based adaptive methods can not apply to analyze extension, dynamic and complexity properties of large-scale video multicast applications. To solve this challenging problem, we develop a hierarchical adaptive architecture for large-scale layered video multicast (HALVM) based on dynamic self-organized agent. HALVM decomposes a large-scale video multicast system into a series of hierarchical sub-systems of small-scale. In HALVM, adaptability is realized by the sources, sinks, and hierarchical agents respectively. HALVM is a promising scheme which can be extended easily and be managed effectively.The central topic of HALVM is how to configure and manage hierarchical agents dynamically. Through analyzing the original topology of video multicast tree, we propose a novel self-organized agent protocol (SL-SOAP) which based on the shared loss Model. In SL-SOAP, we first decompose a logical agent into several processes named OU and OTs respectively, and propose an optimal method to deposit the processes of multiple AVM sessions in multicast network nodes. The control messages and transition statuses of SL-SOAP are defined and designed in details.Scalable video transcoding is the core of HALVM. An ideal transcoding scheme can not only meet the requirement for congestion control, but also optimize truncating bits of enhancement layers according to rate-distortion theory for maximizing receiving utility. We improve the real-time and quality performance of a scalable 3D wavelet coding scheme, MC-EZBC, and then focus on the joint rate-distortion target model of the MC-EZBC transcoding and propose a rate-distortion optimized layered rate control method. Experimental results show that our method can improve the average video quality PSNR by 0.3-0.5dB in comparison with the existing ordered truncating method.Furthermore, in view of the lack of video traffic model which can characterize the statistical...