云环境下网络性能测量与服务优化的研究

发布时间：2018-03-23 11:08

本文选题：网络测量　切入点：云服务　出处：《北京科技大学》2016年博士论文

【摘要】：云计算依托于网络,构建了一种以服务形式分配资源和信息的架构,成为服务科学的一种实现,并极大的满足了各类用户的需求。随着云服务的增加,用户需求的不断增长,云平台及云服务的提供商受到诸多因素,如网络服务质量状况、网络资源分布情况以及云计算服务模式等的影响,在所提供服务的效率和效益上面临诸多挑战。为了解决云用户的服务优化问题,本文首先将云用户分为两类：根据用户所需求服务的特性的不同,将单个用户服务资源消耗相对较少、占用服务资源时间相对较短、服务实时性要求相对较高的一类用户定义为轻量级云用户；将单个用户服务资源消耗相对较多、占用服务资源时间相对较长、服务实时性要求相对较低的一类用户定义为重量级云用户。由于这两类用户服务特性的不同,本文将轻量级云用户存在的服务问题抽象为一个服务请求优化的问题,而重量级云用户则抽象为一个服务部署优化的问题。据此,本文着重考虑对服务的“网络”方面的属性进行优化：通过对现有网络的调研,发现了网络中存在的问题,然后针对问题,对这两类用户分别给出了优化解决方案,进而通过数值仿真实验进行了验证。本文获得的研究成果具有可应用性,将帮助不同层次的云服务提供商在保证用户服务质量的同时,实现资源的优化以及成本的降低,对指导行业的发展具有较大的理论和现实意义。本文所获得的主要研究及创新成果如下：(1)本研究提出了创新性的抓包丢包率预测方法,并纠正了当前学术界对TCP时间戳利用率的错误描述。同时,本研究首次发现了伸展响应现象在互联网上的大量存在,并通过对比实验,深入分析了不同往返延迟测量技术的优劣以及延迟的分布情况；(2)针对第一个研究点所发现的问题,本研究给出了一个云网络整体的优化方案；在此方案下,针对轻量级用户的服务请求优化问题提出了一种优化策略。策略包括一种云服务器负载的预测方法以及相应的服务请求优化算法,最后通过仿真实验证明了负载预测方法和算法的有效性与高效性；(3)本研究深入考虑了重量级云用户的服务部署问题,利用集合论和矩阵的相关数学知识将此问题建模为一个二次分配问题的变形,并提出一种改进型的模拟退火算法进行了服务部署的优化。本研究给出了算法的详细细节并通过数值仿真实验证明了算法的可用性和高效性。
[Abstract]:Cloud computing based on network, construct a distribution in the form of service resources and information architecture, has become a scientific service, and to satisfy the needs of various users. With the increase of cloud services, increasing user demand, cloud platform and cloud service providers by many factors, such as the service quality of the network status, influence of cyber source distribution and cloud computing service mode, faces many challenges in the efficiency and effectiveness of services provided. In order to solve the problem of optimization of the cloud service users, the cloud users are divided into two categories: according to the characteristics of the different needs of the user service, single user service the consumption of resources is relatively small, service resources occupy a relatively short time, and the real-time requirement of a user defined relatively lightweight cloud users; single user service resource consumption is relatively Many service resources occupied a relatively long period of time, a kind of user defined requirements of real-time services is relatively low for the heavyweight cloud users. Because these two types of user service characteristics, the service problems of lightweight cloud users is abstracted as a service request optimization problem, and the user is abstracted as a heavyweight cloud a service deployment optimization problem. Therefore, this paper focuses on the service of "network" attribute optimization: through the investigation of the existing network, find the problems that exist in the network, and then solve the problem, the two kinds of users are given the optimization solution, and through numerical simulation are verified. The research results of this paper can be applied, will help the different levels of cloud service providers to ensure the user quality of service at the same time, optimize resources and reduce the cost of, Has great theoretical and practical significance for guiding the development of the industry. The main research and innovation achievements obtained in this thesis are as follows: (1) this study proposes the prediction method of capture innovative packet loss rate, and correct the current academic circles to describe the TCP timestamp utilization rate error. At the same time, the study found for the first time stretching in response to the phenomenon of the Internet, and through the contrast experiment, analyzes the advantages and disadvantages of round-trip delay measurement and distribution delay; (2) found in the first study of the problem, this paper presents the optimization scheme of a whole cloud network; under this scheme, the lightweight user service request optimization problem and proposes a novel optimization strategy. Strategies include a cloud server load prediction method and the corresponding service request optimization algorithm, finally proved by simulation experiment The effectiveness and efficiency of the load forecasting method and algorithm; (3) this study deeply considered the problem of heavyweight service deployment of cloud users, using mathematical knowledge set theory and matrix of this problem is modeled as a two distribution of deformation, and proposes an improved simulated annealing algorithm is optimized service deployment. This paper presents the details of the algorithm and the numerical simulation experiments prove the availability and efficiency of the algorithm.

【学位授予单位】：北京科技大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP393.06

【相似文献】