基于自律计算的分布式代理系统的设计与实现

发布时间：2018-03-30 17:40

本文选题：自律计算　切入点：分布式代理系统　出处：《哈尔滨工业大学》2014年硕士论文

【摘要】：现如今万维网已经渗透到了人类社会发展的许多方面，成为了不可或缺的一部分。随着网络技术的发展，，各种相应的网络服务也随之出现，然而大规模的网络应用和网络服务使得网络中的流量剧增，导致了网络流量拥塞、网络响应延迟等问题。HTTP请求产生的网络流量所占的比重持续增长，使得Web服务器承担的访问负载越来越大，同时人们对于HTTP请求的响应速度期望也越来越高。代理技术是减轻Web服务器访问负载、提高HTTP请求响应速度的重要措施，而分布式代理技术可以解决单台代理服务器服务能力不足的问题，提供更好的网络服务质量，同时采用了负载均衡技术和缓存技术，提高分布式代理系统的整体性能。为提升缓存服务器的性能，本文提出了以二级域名为基础的内容热度预测算法，为自动灵活的、细粒度的自动管理分布式代理系统，本文建立了基于自律计算的分布式代理系统管理机制，其中运用了自律计算和强化学习的思想。首先，介绍了与本课题相关的理论与技术，了解了MAPE-K自律循环模型的框架结构、组成部分以及相应的功能；了解了当前流行的一些缓存替换算法的原理以及各算法之间的优缺点有了更加深入的理解；通过举例理解了Q学习算法的原理以及简单应用。其次，讲述了基于自律计算的分布式代理系统的设计，提出了分布式代理系统框架结构，包括负载均衡、代理以及缓存等三个模块；提出了自律的分布式代理系统模型，从而将自律计算技术与分布式代理系统结合在一起；提出了基于二级域名的缓存内容热度预测方法，不仅考虑了缓存对象的命中次数，还考虑了缓存对象的整体特征，将热度较高的URL迁移出去，从而降低了缓存服务器的负载。然后，介绍了基于自律计算的分布式代理系统的实现，重点介绍了基于Q学习的自律决策模块，提出了基于BP神经网络的Q学习模型和算法。最后，通过基于Q学习的自律决策框架模型和缓存内容热度预测实验结果，表明了分布式代理系统的自律管理机制能够有效的、灵活的实现系统管理，并提升了系统的整体性能。
[Abstract]:Nowadays, the World wide Web has permeated many aspects of the development of human society and become an indispensable part. With the development of network technology, various kinds of corresponding network services have also appeared. However, large-scale network applications and network services make the traffic in the network increase dramatically, resulting in network traffic congestion, network response delay and other problems. The proportion of network traffic generated by HTTP requests continues to grow. It makes the Web server bear more and more access load and people expect the response speed of HTTP request more and more. Proxy technology is an important measure to lighten the access load of Web server and improve the response speed of HTTP request. Distributed proxy technology can solve the problem of insufficient service capacity of a single proxy server and provide better network quality of service. At the same time, load balancing technology and cache technology are adopted. In order to improve the performance of cache server, this paper proposes a content heat prediction algorithm based on secondary domain name, which is an automatic, flexible and fine-grained distributed agent management system. In this paper, a distributed agent system management mechanism based on autonomous computing is established, in which the idea of autonomous computing and reinforcement learning is used. Firstly, the theory and technology related to this topic are introduced, and the frame structure, components and corresponding functions of the MAPE-K autonomous cycle model are understood. The principle of some popular cache replacement algorithms and the advantages and disadvantages of each algorithm are understood more deeply. The principle of Q learning algorithm and its simple application are understood by examples. Secondly, This paper describes the design of distributed agent system based on autonomic computing, puts forward the framework of distributed agent system, including three modules of load balancing, agent and cache, and puts forward the model of autonomous distributed agent system. Combining autonomous computing technology with distributed agent system, this paper proposes a new method for predicting cache content heat based on secondary domain name, which not only considers the hit times of cache objects, but also takes into account the overall characteristics of cache objects. In order to reduce the load of cache server, the URL with high heat is migrated out. Then, the realization of distributed agent system based on self-discipline computing is introduced, and the autonomous decision-making module based on Q-learning is introduced emphatically. The Q learning model and algorithm based on BP neural network are proposed. Finally, through the self-discipline decision framework model based on Q-learning and the experimental results of cache content heat prediction, it is shown that the autonomous management mechanism of distributed agent system is effective. Flexible implementation of system management, and improve the overall performance of the system.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.09

【共引文献】