基于协同演化算法的云计算资源调度的研究

发布时间：2018-05-28 12:46

本文选题：云计算 + 协同演化　；参考：《电子科技大学》2014年硕士论文

【摘要】：云计算作为使用超大规模的虚拟化资源为用户提供动态服务的方式,受到了学术界和工业界的广泛关注和研究。云计算平台将各种资源整合到一个虚拟资源池中,云计算的用户按照他们不同的需求从资源池中获得不同的计算资源、存储资源以及软件服务。如何对于资源池各种资源进行调度、管理以及其调度效率的优劣直接影响整个云计算平台的工作性能和表现,并且云计算为各种用户提供服务,而这些用户期望达到的目标也不尽相同,就导致了云计算平台的资源调度算法的不同。本文对云计算的架构、商业模型进行了阐述,并对云计算的调度算法的研究现状进行了全面分析,针对现有调度算法中存在的一些不足,主要进行了以下几个方面的研究工作:(1)提出了一种改进的合作型的协同演化算法用于解决云计算环境下的多目标调度问题。用户使用云计算平台执行任务时,需要租借云计算平台中虚拟机资源,产生执行成本,而用户又希望任务能够尽快的执行完成,但是执行成本和完成时间又是相互冲突的两个目标,只能取得这两个目标的均衡结果。在已有的云计算研究中,通常利用加权方法把二者转换成为单目标进行调度的研究,每一次只能为用户提供一种调度方案,并且很多研究都忽略任务之间的依赖关系。因此,本文对具有时序约束关系的任务调度进行建模,并针对成本和完成时间的多目标问题,提出了具有以田口方法为基础进行个体的定向变异的微观操作以及多种群和外部最优解种群之间的合作进行全局空间搜索的宏观操作的改进的合作型的协同演化调度算法。(2)在以往的云计算调度算法的研究中,通常把任务之间的数据传输的所造成的影响进行忽略,但是数据传输和数据的本地性对数据密集型的任务有很大的影响,尤其是对于科学工作流调度具有更大的影响。因此,本文对数据传输和数据本地性的重要性进行了说明,并针对数据密集型的任务构成的科学工作流的调度问题,考虑到科学工作流的用户更加关心的是完成时间,提出了以最小化完成时间为目标,结合了免疫算法和协同演化算法的优势的竞争型协同免疫算法用于解决科学工作流的调度问题。(3)利用Cloud Sim搭建云计算仿真平台,对本文提出的两种调度模型和对应的算法进行实验,试验结果验证了本文提出的两种调度算法的有效性。
[Abstract]:Cloud computing, as a way to provide dynamic services to users using large scale virtualization resources, has received extensive attention and research from academia and industry. Cloud computing platform integrates all kinds of resources into a virtual resource pool. Cloud computing users obtain different computing resources, storage resources and software services from the resource pool according to their different needs. How to schedule, manage and schedule all kinds of resources directly affects the performance and performance of the whole cloud computing platform, and cloud computing provides services for all kinds of users. These users expect to achieve different goals, which leads to different resource scheduling algorithms of cloud computing platform. In this paper, the architecture and business model of cloud computing are described, and the research status of cloud computing scheduling algorithm is comprehensively analyzed. In this paper, we propose an improved cooperative co-evolution algorithm to solve the multi-objective scheduling problem in cloud computing environment. When users use cloud computing platform to execute tasks, they need to rent virtual machine resources in cloud computing platform to generate execution cost, and users hope that tasks can be completed as soon as possible. However, execution cost and completion time are two conflicting objectives, which can only achieve balanced results. In the existing cloud computing research, the weighted method is usually used to convert them into single-objective scheduling research. Each time, only one scheduling scheme can be provided for users, and many researches ignore the dependencies between tasks. Therefore, this paper models the task scheduling with temporal constraints, and aims at the multi-objective problem of cost and completion time. In this paper, an improved cooperative co-evolution with micro-operations based on Taguchi method for individual directional variation and cooperation between multiple groups and external optimal solution populations for global spatial search is proposed. In the past research of cloud computing scheduling algorithm, The impact of data transmission between tasks is usually ignored, but data transmission and data nativeness have great influence on data-intensive tasks, especially on scientific workflow scheduling. Therefore, this paper explains the importance of data transmission and data nativiality, and considers that the users of scientific workflow are more concerned with the completion time in view of the scheduling problem of scientific workflow constituted by data-intensive tasks. A competitive cooperative immune algorithm, aiming at minimizing the completion time and combining the advantages of immune algorithm and co-evolution algorithm, is proposed to solve the scheduling problem of scientific workflow. It uses Cloud Sim to build a cloud computing simulation platform. Two scheduling models and corresponding algorithms proposed in this paper are tested and the experimental results verify the effectiveness of the two scheduling algorithms proposed in this paper.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.01

【相似文献】