Optimization Research For Internet Content Distribution

With the rapid increase of multimedia services and Internet users, Internet Content Providers have to face enormous requests from gigantic user groups. It is hard for traditional Client/Server based infrastructure to meet these challenges: servers' power and bandwidth of their access points are always the bottlenecks of system scalability and performance. This situation calls for new technologies.Two dedicated content distribution networks emerged to take the role of content servicing. One is peer-to-peer networks: users in the Internet form an overlay network and cooperatively distribute/service the contents. The most important characteristic of P2P is direct resources sharing between users, hence solve the bottlenecks of servers' scalability and bandwidth limitations. The key technology is location algorithm for distributed objects. Another is Content Delivery Networks, which is deployed/maintained by commercial operators. CDN networks push contents to the edge of the networks, hence reduce core network traffic at the same time reduce user perceived response delay. There are many challenges in the research of content distribution networks. And there are still spaces for further improvement for many algorithms, as optimization theory and technology are not well exploited yet.The works of this dissertation concentrate on optimized design of some aspects of content distribution networks, including: optimized resource query algorithm design in unstructured P2P networks; online replica placement scheduling in CDN networks; cooperation between P2P and CDN networks.The contributions of the dissertation are:(1)Design defects of previous controlled-flooding algorithms are analyzed first; based on optimization principal, a novel Selective Dynamic Query algorithm is proposed. This algorithm can achieve the same traffic control level as previous Dynamic Query algorithm, together with approximate response delay as Expanding Ring. By extensive experiments of different topologies and scenarios, SDQ always demonstrates superior performance than other query algorithms.(2)To meet the online scheduling requirement of CDN replica placement problem, a traditional mathematical model can't be solved in a relatively short time. A feasible replica placement framework is presented. In the system level, the number of variables and constraints are greatly reduced by clustering. A hierarchical model is introduced in: the problem is decomposed to inter-section level schedule and inner-section level schedule, each with its own mathematical programming, this layered abstraction greatly reduces problems dimensions. The structure of the obtained formulation is re-optimized. After all these steps, the computational burden is reduced by magnitude. We all apply this framework for a real CDN media scheduling case.(3)BitTorrent-like overlay P2P applications neglect underlying network topologies, hence unnecessarily increase traffics in both ISPs' core networks and inter-ISPs interfaces. This research propose to further explore topological and geographical approximation of peers, and two algorithms are proposed. The first algorithm improves ISP-biased neighbor selection by take access network neighborhood into consideration; the second algorithm proposes that assisted by CDN, a node can choose neighbors from those peers which not belong to same ISP while within the same city, to improve traffic pattern. Experiments demonstrate that these proposals can greatly reduce traffics on both ISPs' core networks and inter-ISPs interfaces.The achievements of the dissertation can be used to promote the development of content distribution networks.