Research On The Stability Of Media Streaming Distribution In The Next Generation Internet

With the rapid development of Internet services and the widely spreading of media streaming technology, more and more applications press urgent demands on the grouping communication pattern. Grouping communication can be realized by unicast, IP layer multicast and Application-Layer Multicast (ALM). Due to the drawbacks of traditional unicast and IP multicast distribution patterns, ALM has attracted more attention from researching area and became the most popular solution for the data transmission of grouping communication. The Next Generation Internet (NGI) provides various access methods for mass subscribers which intensifies the heterogeneity of terminal performance and users' dynamic characteristics. Although seems to be the most efficient and advantageous technique for the distribution of larger-scale media streaming, ALM still faces lots of new technical challenges.ALM employes end-users'terminals to form an overlay network and relay media packets for the purpose of grouping distribution, which means that the dynamic characteristics of hosts may directly affect the performance of data transmission. Considering the tree-based topology, the leave or failure of any relaying node may results in the disconnection of its offspring nodes, not to mention the guarantee of continuity or timeliness for streaming media. This phenomenon is concluded as the stability of application-layer multicast which has strong relationship with the overall performance and QoS of the provided media services, even quality of experience from the perspective of end-users in real-time application scenario such as live media streaming. Therefore, solving the problem of ALM stability is the important precondition for deploying large-scale media streaming services in next generation Internet. This dissertation focuses on the key techniques of improving the stability of application-layer multicast. Based on some shortcomings in current techniques, we make an in-depth researching on the possible methods on the enhancement of ALM stability. The main contributions of our work include:(1). Generalize the corresponding relationships and mapping models among cross-layer parameters in the tree-based ALM topology. According to the existing evaluation metrics and organization structure of live streaming applications, we first classified the multiple evaluation metrics into different layers, namely network-layer, application-layer and P2P-layer respectively. Based on the categories of candidate parameters, a mapping matrix which indicates the potential relationships and coordinated control of ALM performance among cross-layer metrics is proposed for furthering measurement. We modeled some important corresponding relationships by using multiple traffic collecting and measuring methods in our campus networks and streaming application testbed. This work defines our research directions of later peer selection mechanism and the decision of QoE assessment model.(2). In order to reduce the affected scope of node leaving or failure, we proposed a Topology Information and Multiple Attributions Decision based Peer Selection Mechanism (TIMAD-PSM) according to the analysis of cross-layer parameters mapping models. The problem is further divided into two sub-problems as P2P-layer and network-layer priorities determination. We carried out a Centralized Peer Selection Algorithm (C-PSA) to solve this problem by an entropy weighting based multiple attributions decision model and linear weighting based integrated priority index. For the purpose of further validating the enhancement of users'experience by employing C-PSA, we proposed a novel objective reference-aware QoE assessment index and provided the definition of the QoE index with a tuple representation. The experiment results prove that our C-PSA would efficiently reduce the affecting range of node departuring behavior. Meanwhile, the QoE evaluation model is effective and deployable.(3). For the application recovery and topology reconstruction, we proposed a Global-Optimization Reactive Reconstruction Algorithm (GO-R2A). In our issue, we first modeled the reconstruction scenario into Degree and Hop Constrained Maximum number of Affected Nodes Spanning Tree (DHC-MANST) problem and presented a Prim-similarity algorithm to improve the overall performance and balance of recovery topology with the sacrifice of a little acceptable operation delay. We also evalated some useful metrics such as the depth of reconstruction tree, successful recovery rate and the topology reconstruction delay as the indicators of the recovery performance profits and cost by simulation. The statistical results demonstrate that GO-R2A keeps consistent with the design expectation which is helpful to the enhancement of overall reconstruction performance and the stability of ALM.(4). Designing and preliminarily implementing the architecture of large-scale application distribution in next generation Internet. This dissertation combined the various data transmission patterns offered by CDN and P2P techniques and raised a clustering and hierarchical architecture to support the heterogeneous access methods in NGI based on the feasibility of proxy ALM. At the meanwhile, we clearly defined the functions of each module in the architecture and the procedure of participating into application services, which provides the guiding opinion for PeerCast prototype modification in our testbed and the preliminary proof of deployability from the practical point of view.