弥散云资源感知与调度方法研究与实现

发布时间：2018-03-25 17:00

本文选题：弥散云资源　切入点：感知　出处：《河北经贸大学》2014年硕士论文

【摘要】：云计算是一种实现大规模计算的信息处理方式，本质上是利用虚拟化技术、分布式计算技术和网络技术将分散的云基础单元（简称云元）汇聚到一起形成共享资源池，并以按需、弹性和可度量的方式向用户提供云服务，这些“云元”通常跨集群、室、甚至数据中心分布并随时间动态变化，随着分散的“云元”在时空上动态变化，资源池中的海量资源呈现弥散特征，形成弥散云资源，即一种分布式的具有时变性的变粒度的云资源。怎样在海量资源中准确快速感知并调度这些资源，提供低成本高性能可伸缩的云服务，提高用户满意度是目前和今后云计算技术领域面临的重要问题。本文从IaaS资源提供方和请求方两个角度，研究弥散云资源的感知与调度问题，主要内容有：（一）从IaaS资源提供方角度，提出基于MA感知的弥散云资源调度方法，主要研究：（1）提出基于移动Agent范型的IaaS层资源部署架构，并给出弥散云资源感知方法：基于作业完成时间分布预测，通过设置置信区间，剔除不符合条件的策略，一次性压缩策略空间，引入Skyline计算思想细粒度抽取价值资源，形成资源视图；（2）研究基于移动Agent感知的弥散云资源调度方法MA_RS：利用基于移动Agent的资源感知层感知的信息，利用弥散特性，结合多任务的松耦合特性，构建“搜索——决策——执行”这样一个多阶段迭代调度模型，实时细粒度的捕获工作负载的局部特性，并根据相邻阶段工作负载的相似性特点，在工作负载变化剧烈（平缓）处自适应动态分割（合并）调度区间，实现整体调度性能的提升；（二）基于移动Agent的MapReduce云计算计算框架，针对云资源弥散性，从资源请求方的角度考虑提出了一个公平共享指标来实现高性能和公平性的基于云资源弥散性感知的公平调度方法，，主要研究：（1）构建基于移动Agent的MapReduce分布式计算框架MapReduce_MA，定义移动Agen（t如Master_MA和Slaver-_MA）的功能集合，并具体实现了移动Agent的主要功能；（2）研究基于云资源弥散性感知的公平调度方法，引入“共享进度份额”来定义共享和公平，根据用户偏好选择任务，依据性能函数匹配相应资源，并给出了基于云弥散性感知的公平调度算法。这种调度方法不仅权衡了成本和效益，而且能够在异构集群中提供良好的性能和公平性。本文将上述架构和方法在惠普实验室开放Cirrus集群上进行了有效性评估并通过实例验证的形式说明了本文感知调度方法的潜在好处。
[Abstract]:Cloud computing is a kind of information processing method to realize large-scale computing. In essence, it uses virtualization technology, distributed computing technology and network technology to bring together scattered cloud base units (cloud elements) to form a pool of shared resources. And provide cloud services to users on demand, flexibility, and measurability. These "cloud elements" are typically distributed across clusters, rooms, and even data centers, and vary dynamically over time, as dispersed "cloud elements" change dynamically in time and space. The massive resources in the resource pool present the characteristics of dispersion and form the diffuse cloud resources, that is, a kind of distributed, time-varying and variable granularity cloud resources. How to accurately and quickly perceive and schedule these resources in the mass resources? Providing low cost and high performance scalable cloud services and improving user satisfaction are important problems in cloud computing technology field at present and in the future. In this paper, we study the perception and scheduling of diffuse cloud resources from the perspective of IaaS resource provider and requester. The main content of this paper is: (1) from the point of view of IaaS resource provider, we propose a MA aware distributed cloud resource scheduling method. In this paper, we propose a IaaS layer resource deployment architecture based on mobile Agent paradigm, and present a distributed cloud resource awareness method: based on the prediction of job completion time distribution, by setting confidence interval, we can eliminate the non-conforming strategy. In this paper, we introduce the idea of Skyline computing to extract value resources in a one-off compressed policy space, and form a resource view. (2) to study the distributed cloud resource scheduling method based on mobile Agent perception, MARSs: using the information of resource awareness layer based on mobile Agent. A multi-stage iterative scheduling model named "search-decision execution" is constructed by using the dispersion property and the loose coupling of multi-task. The real-time fine-grained local characteristics of the workload are captured. According to the similarity characteristics of workload in adjacent phases, adaptive dynamic partition (merging) scheduling interval is realized at the place where workload changes dramatically (flat), so as to improve the overall scheduling performance. (2) MapReduce cloud computing framework based on mobile Agent. In view of the dispersion of cloud resources, this paper proposes a fair sharing index to achieve high performance and fairness, which is based on the knowledge of cloud resources diffusion and fair scheduling. This paper mainly studies how to construct MapReduceMA-based MapReduce distributed computing framework based on mobile Agent, define the functional set of mobile Agen(t such as Master_MA and Slaver-Stack, and implement the main function of Mobile Agent.) A fair scheduling method based on cloud resource diffusion sexy knowledge is studied. "share schedule share" is introduced to define sharing and fairness. According to user preference, tasks are selected and corresponding resources are matched according to performance function. A fair scheduling algorithm based on cloud diffusion sexy knowledge is presented, which not only balances cost and benefit, but also provides good performance and fairness in heterogeneous clusters. This paper evaluates the effectiveness of the above architecture and methods on HP Labs open Cirrus cluster and illustrates the potential benefits of this method by example verification.
【学位授予单位】：河北经贸大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.01

【引证文献】