基于分布式缓存的内容分发网络

发布时间：2018-05-16 22:04

本文选题：分布式缓存 + 分布式传输　；参考：《西安电子科技大学》2014年硕士论文

【摘要】：本文提供了一种新的内容分发网络的架构，即通过在网络中部署具有低成本、大容量的缓存节点来减轻用户下载文件对网络回传容量的造成的开销，但也引入了内容文件的缓存、传输等问题，加剧了系统的复杂度。因此对内容分发过程的高效管理在内容分发网络中变的迫切且重要。本文首先分析了内容分发网络中的分布式缓存问题和分布式传输调度问题，并且给出了针对这两种问题的常见算法。然后研究了网络编码在基于分布式缓存的内容分发网络中的应用。即通过使用网络编码可以将文件的缓存问题和传输问题转化成使网络为每个移动用户保证一个相对固定且足够大的流速率的问题，从而为内容文件的分发问题建立了一个流调度的模型。利用原始-对偶的分解方法对流调度问题进行分解可以得到中心服务器与helper节点之间的缓存分配问题和helper节点与移动用户之间的带宽分配问题。通过将原问题进行分解，每一个网络节点可以获得一个可以分布式运行的子算法，各个子算法之间联合迭代，最终得出最优的缓存、带宽分配。最后，考虑到移动无线环境动态变化的马尔科夫特性，本文中将整个网络拓扑的变化看为一个有限状态的马尔科夫链，，每一种网络拓扑都是这个马尔科夫链的一个平稳分布。根据马尔科夫状态的平衡方程可以得到网络拓扑的状态转移概率和转移周期。运行分布式算法的节点可以根据这个状态转移概率和周期来进行自我配置从而使整个网络的性能达到全局最优。
[Abstract]:In this paper, a new architecture of content distribution network is proposed, in which the cost of downloading files to the network is reduced by deploying a low-cost, large-capacity cache node in the network. However, some problems such as cache and transmission of content files are introduced, which aggravate the complexity of the system. Therefore, the efficient management of the content distribution process becomes urgent and important in the content distribution network. In this paper, we first analyze the distributed cache problem and distributed transport scheduling problem in content distribution networks, and give some common algorithms for these two problems. Then the application of network coding in distributed cache-based content distribution network is studied. That is, by using network coding, the problem of file caching and transmission can be transformed into the problem of making the network guarantee a relatively fixed and sufficiently large flow rate for each mobile user. Thus, a flow scheduling model is established for the distribution of content files. The problem of cache allocation between the central server and the helper node and the bandwidth allocation problem between the helper node and the mobile user can be obtained by decomposing the primal-dual decomposition method for the convection scheduling problem. By decomposing the original problem, each network node can obtain a distributed sub-algorithm that can be run, each sub-algorithm iterates together, and finally the optimal cache and bandwidth allocation are obtained. Finally, considering the Markov characteristic of the dynamic change of mobile wireless environment, the change of the whole network topology is regarded as a finite state Markov chain in this paper, and each network topology is a stationary distribution of the Markov chain. According to the equilibrium equation of Markov state, the state transition probability and transition period of network topology can be obtained. The nodes running the distributed algorithm can self-configure according to the transition probability and period of the state so that the performance of the whole network can be optimized globally.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.02

【参考文献】