云环境下资源调度策略的研究与分析

发布时间：2018-09-12 18:43

【摘要】：自云计算概念被提出后,便迅速引发一场全球性的研究和开发热潮,很多新兴的云产品被纷纷推向市场。云计算是继分布式处理、并行处理、网格计算、公用计算和虚拟化计算之后的一种按需来提供服务的计算模式。云计算利用虚拟化技术将数据中心底层物理资源抽象成虚拟单元,根据用户的需求为其提供各类服务。为适应用户日益增加的需求,数据中心作为云计算的重要载体也在高速的膨胀,逐渐增加的IT设备使得云资源提供商及企业不得不考虑数据中心能耗、资源利用率、系统性能、服务质量、负载、成本控制等问题。目前,云计算在资源调度方面的研究有几个主要方向:以提高性能为目标的资源调度,以提高服务质量为目标的资源调度,以经济原则为中心的资源调度,以提高资源利用率为目标的资源调度,以降低能耗为目标的资源调度,以平衡负载为目标的资源调度。云系统中的资源类型有很多,比如CPU,内存,硬盘,网络带宽,I/O设备等,且系统中的资源是动态变化的。用户向云端提交任务,各个用户提交的任务数量是不同的,且每个任务对资源的需求类型以及对每种类型资源的需求数量也是不同的。云系统中用户和云提供商双方利益角度不同,一个好的资源调度策略需要均衡各个参与者的利益。本文设计的基于偏好的公平分配策略FABP(Fair allocation strategy based on preference),给出了用户优先级和任务优先级的定义,采用稳定匹配理论来解决虚拟机置放问题。该算法在进行任务调度时先通过用户优先级定位到某个用户,再根据任务优先级选择该用户相应的任务进行处理。在资源分配方面,先建立任务和物理机之间的偏好关系,再运用稳定匹配理论将任务置放到最合适的物理机上,以减少资源碎片的产生。实验结果表明,该算法不仅能缩短平均任务调度时间,而且保证了任务调度过程中用户和任务的公平性,从而提高了用户的满意度,也实现了综合资源利用率的最大化,从而保证了云资源提供商的利益。
[Abstract]:Since the concept of cloud computing was put forward, it has triggered a global research and development boom, and many new cloud products have been launched into the market. Cloud computing is an on-demand computing model following distributed processing, parallel processing, grid computing, common computing and virtualization computing. Cloud computing abstracts the underlying physical resources of the data center into virtual units using virtualization technology and provides various services according to the needs of users. In order to meet the increasing demand of users, data center as an important carrier of cloud computing is expanding at high speed. With the increasing of IT devices, cloud resource providers and enterprises have to consider data center energy consumption, resource utilization, system performance. Quality of service, load, cost control and so on. At present, the research on resource scheduling in cloud computing has several main directions: resource scheduling aiming at improving performance, resource scheduling aiming at improving quality of service, and resource scheduling centered on economic principle. Resource scheduling aims at improving resource utilization, resource scheduling aiming at reducing energy consumption, and resource scheduling aiming at balancing load. There are many resource types in cloud systems, such as CPU, memory, hard disk, network bandwidth, I / O devices and so on, and the resources in the system are dynamic. Users submit tasks to the cloud, and the number of tasks submitted by each user is different, and the types of requirements for resources and the number of requirements for each type of resources are different for each task. The interests of users and cloud providers are different in cloud systems. A good resource scheduling strategy needs to balance the interests of each participant. In this paper, we design a fair allocation strategy based on preference, FABP (Fair allocation strategy based on preference), which gives the definitions of user priority and task priority, and uses the theory of stable matching to solve the problem of virtual machine placement. The algorithm firstly locates a user through the priority of the user and then selects the corresponding task according to the priority of the task to process the task. In the aspect of resource allocation, the preference relationship between task and physics machine is established first, then the task is placed on the most suitable physical machine by using the theory of stable matching, so as to reduce the occurrence of resource fragment. Experimental results show that the algorithm can not only shorten the average task scheduling time, but also ensure the fairness of users and tasks in the task scheduling process. Thus ensuring the interests of cloud resource providers.
【学位授予单位】：江西理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP3

【参考文献】