Research On Internet-Oriented Fast Data Distribution Technology Under Timing Constraint

Data distribution under timing constraint has been widely applied in the Internet,such as live video streaming, video on demand, video conference, online music, VoIP anddistance learning. The organization and scheduling of dynamic resources are the key is-sues to the large-scale distribution for time series data. Since the Internet resources aredistributed, dynamic and autonomous, it is hard to utilize them efficiently. Peer-to-Peertechnology provides a valid way to make use of the Internet resources, but the researcheson P2P protocol mainly focus on the top-down design to integrate system, so it is hard toreuse the service of resource organization and resource location to assist the developmentof up-layer applications. At the same time, the existing distribution and scheduling ap-proaches can not efficiently utilize the Internet resources, which leads to the performancereduction on fast data distribution. In doing so, this dissertation aims at the effective uti-lization of the dynamic Internet resources and studies the issues on fast data distributionunder timing constraint. The main contributions are listed as follows:(1) The models on organization and allocation of Internet resources. To efficientlyorganize and utilize the dynamic Internet resources, the dissertation proposes an organi-zation model and an allocation model, provides some supporting technologies for system,and allows the application developers to focus on the designs of application logic.(2) The requests-matching algorithm based on the Bipartite Graph. To maximize theutilization of service resources, the dissertation defines the optimal problem on through-put scheduling and then proposes a chunk scheduling algorithm based on the matching ofBipartite Graph, namely BBS, which can improve the quality of streaming transmissionandtheperformanceonstartupdelay. ThemainideaofBBSisto(i)transformtheoptimalproblem on throughput scheduling into finding the maximum matches in the correspond-ing Bipartite Graph and then (ii) assign the chunk requests to each serving peer, whichachieve the aims to maximize the utilization ratio of serving resources and optimize thethroughput of transmission.(3) The services-scheduling algorithm for fast data distribution under timing con-strain. Due to the fact that it is hard to identify the most urgently needed chunk for systemin the view of requesting peers, the dissertation proposes a services-scheduling algorithm for Deadline-Sensitive Fast data distribution: namely DSF. The main idea of DSF is thatthe serving peer evaluates urgency of a chunk according to the deadline and the numberof requesting peers and then gives priority to transfer the most urgently needed chunk tothe requesting peer with the higher service capacity, which achieve the aim to acceleratethe data distribution of the system under the timing constraint-satisfaction.(4) The design of an extensible peer-to-peer simulator. To study the technologieson requests-matching and services-scheduling, this dissertation designs and implementsa flexibly extended peer-to-peer streaming simulator: EPSS. EPSS can flexibly extendthe functions so as to support (i) the strategies on matching requests, scheduling servicesand building overlay and (ii) the simulations in the dynamic network environment. Themethodology of EPSS is to analyze the structure of the existing P2P simulators and thenfindouttheirdisadvantagesandthecauses. Bytheclearmodulepartitionsandfunctionallyextensible designs, it achieves the aims to efficiently support many kinds of studies andextend functions.(5) The applied and experimental evaluation based on iVCE platform. The disserta-tion integrates (i) the models on organization and allocation of Internet resources, (ii) therequests-matchingalgorithmbasedontheBipartiteGraphand(iii)theservices-schedulingalgorithm for deadline-sensitive fast data distribution and applies them into iVCE plat-form. To do so, we implement and deploy a streaming prototype system based on iVCEplatform. The experiments validate our claims on the models and algorithms.