【作者】 王玮；

【导师】 刘文予；

【作者基本信息】 华中科技大学 ， 通信与信息系统， 2009， 硕士

【摘要】 内容分发网络是为了改善网络负载不均,增加用户满意度而发展起来的热门网络技术。目前最典型的内容分发网络包括点对点网络(P2P)与内容发布网(CDN)两种,文章首先简要介绍了点对点网络与内容发布网的概念和发展现状,并详细给出了内容发布网网络中的几种主要技术:内容路由、内容分发、内容管理。本文的主要工作围绕商业应用中的PUSH模型展开。它与前人研究的PUSH技术相比,有两个特点。一是更高的时效性:至少在一个触发时隙内能完成求解;二是求解的连续性。与一般的内容分发模型相比,本研究与前人文章主要有四点不同:一是简化拓扑到以节点为单位。把一片代理区域内所有用户和代理服务器看做一个统一节点,不考虑具体的边缘服务器到用户的路由,这样做极大的提高了模型规模的可扩展性,而且还把具体的路由技术从内容分发技术中剥离开。二是确立了以内容发布网运营商为服务对象,完全从内容发布网运营商的角度出发考虑问题,不仅使模型的目标函数和限制条件更完备,也更有实际说服力。三是采用内容聚类的思想,把多个热度相似的内容组合成一个频道,在调度时看成一个整体。四是在求解多目标函数的模型时,本文没有把两个目标放在一起加权求和,而是假设两种应用环境,分成两个子模型分别求解。最小化网络带宽问题的数学模型是一个标准0-1规划模型,本文不仅用标准的隐数法求解此模型,还给出了自己的启发式解法,并通过大量实验表明,在线性求解时间内,启发解的性能达到了最优解的95%。而最小化服务器损耗问题的数学模型属于混合整数规划,变量数目很多时根本无法求最优解。本文用启发式算法求解,并用估计的最优解的上限与之对比,以反映启发解的性能。实际仿真时做出了两个小的改进,通过对比实验表明,性能比以前调高10%左右。更多还原

【Abstract】 Content Distribution Network is a hot technology aimed to improve the network load balance, increase customer satisfaction. The most typical content delivery network are P2P and CDN. We have given a brief introduction to both P2P and CDN first, then in detail discussed about three technology in CDN: content dilivery, routine, and management.This thesis mainly studies one type of content delivery strategy for CDN, named realtime-PUSH strategy. Compare to normal PUSH strategy, it has two characters: first, it must be much more efficient in time; second, the environment is a continuing process.Compared to former research, our model has four differences: 1, We have brought node clustering structure to our model, which means we no longer care about the different behaves between proxy, client , or source, but combine them as a whole node. In this way, the scalability of model is greatly improved. 2, The main purpose of this research is to help CDN provider, so that the model has precise goals and limits, and is much closer to reality. 3, All the contents are clustered as some bucks, called channel. 4, When facing a extremely complicate problem, we depart the problem into two sub-problems, and then gives two different model to fix out each.The goal of the first model is to minimize the bandwidth consuming. This model has turned out to be a standard 0-1 linear programming. Later in this thesis, we not only give the standard method to solve this problem, but also give a heuristic algotithm. Through a huge number of experiments, we prove that the heuristic algorithm is much more efficient in time and its solution is about 95% close to the optimal one. The goal of the second model is to minimize the exchange of all channels on all proxies. This model proved to be a mixture nonlinear programming problem, and is very hard to get its optimal solution. So we have to give a heuristic algorithm instead. In order to estimate the perfomence of this heuristic algorithm, also give a estimation on the optimal solution. At last made two small changes on the algorithm, under certurn circumtances, the performance can be improved by 10%, with the price of 2 or 3 times of time consuming.更多还原