基于遗传算法的面向能耗的云工作流调度优化

发布时间：2018-03-26 05:24

本文选题：能耗　切入点：工作流　出处：《浙江工商大学》2017年硕士论文

【摘要】：自2006年底IT巨头公司——谷歌公司推出了 "谷歌101"计划,许多知名的IT公司纷纷效仿,IBM、亚马逊、微软、雅虎等陆续展开了自己的"云"计划,其中,IBM推出了 "蓝云"计划,亚马逊推出了"弹性云"计划、中国移动推出了 "大云"计划,阿里巴巴推出了"集团云"计划,甚至,一些不知名的小公司也纷纷运用云计算的模式。云计算已经成为各公司竞相追逐的热点,未来的发展空间非常开阔。云计算的快速发展带动了云计算数据中心在世界各地的广泛建立,但是,这些大规模数据中心的能耗高、污染高的问题已经引起了人们的关注。当前对能耗的管理优化的研究,主要通过动态调整服务器CPU的电压/频率及采用虚拟化技术(包括虚拟机的移植、分配、合并等)、独立并行任务的重组(包括合并集中和迁移等方法)减少所需服务器端数量、关闭空闲服务器或使其处于睡眠状态等方式来降低能耗,很少考虑采用基于工作流的调度优化技术。当前大多数包括遗传算法在内的云工作流执行/调度优化方法通常只关注执行时间或成本的优化,很少考虑能耗因素,少数基于DVFS技术的云工作流执行/调度优化方法考虑了能耗因素,但DVFS技术需要调整服务器CPU的运行电压/频率、反复关闭/重启服务器,这会影响服务器的性能,也会加大服务器组件的磨损率,进而引发服务器采购与替换成本的上升,在实际使用中有一定的局限性。因此,创造性地开展云计算环境下面向能耗的工作流调度优化研究,保持甚至提高云计算服务质量的同时,降低云计算数据中心的能耗,打造绿色云服务系统有重要的科学意义。针对当前缺乏有效的面向能耗的云工作流调度优化方法,研究建立了面向能耗的云工作流过程模型、资源模型,提出了基于负载的能耗计算方法和面向能耗的云工作流调度优化算法。本文采用的算法为基于遗传算法的云工作流分组调度算法(GWSGA算法),相对于现有的关注于时间或成本的调度优化方法,考虑了能耗因素。GWSGA算法编码方式为首先对工作流进行分组,确定工作流中每个任务的组别,并始终把任务组别小的放在组别大的前面,以确保个体编码的有效性,接下来根据工作流的组别、任务要求和资源可得性得到初始种群,这种方式可以有效防止出现不可行解,所以GWSGA算法最大的优点是初始种群的所有解都是可行的,该算法搜索时间非常短、效率非常高,广泛适用于各种科学工作流。仿真实验表明,本文采用的GWSGA算法不仅可以确保云工作流执行的时间效率,而且可以有效降低主机处理任务所消耗的能耗。
[Abstract]:Since Google, the IT giant, launched the "Google 101" project at the end of 2006, many well-known IT companies have followed the example of IBM, Amazon, Microsoft, Yahoo, etc., one after another launched their own "cloud" projects, among which IBM launched the "Blue Cloud" program. Amazon launched the flexible Cloud, China Mobile launched the Big Cloud, Alibaba launched the Group Cloud, and even, Some small, unknown companies are also using the cloud computing model. Cloud computing has become a hot competition among companies. The future is very open. The rapid development of cloud computing has led to the widespread establishment of cloud computing data centers around the world, but these large-scale data centers have high energy consumption. The problem of high pollution has attracted people's attention. The current research on energy management optimization is mainly by dynamically adjusting the voltage / frequency of the server CPU and using virtualization technology (including the transplantation and distribution of virtual machines). Merging, reorganizing independent parallel tasks (including consolidation, migration, etc.) reduces energy consumption by reducing the number of server-side requirements, shutting down idle servers, or keeping them in sleep. Currently, most cloud workflow execution / scheduling optimization methods, including genetic algorithm, usually focus on the optimization of execution time or cost, and rarely consider the energy consumption factor. A few cloud workflow execution / scheduling optimization methods based on DVFS technology take into account the energy consumption factor, but DVFS technology needs to adjust the running voltage / frequency of the server CPU, shut down / restart the server repeatedly, which will affect the performance of the server. It will also increase the wear rate of server components, which will lead to the increase of the cost of server purchase and replacement, which has some limitations in practical use. Therefore, the research of workflow scheduling optimization for energy consumption in cloud computing environment is carried out creatively. While maintaining or even improving the quality of cloud computing services, it is of great scientific significance to reduce the energy consumption of cloud computing data centers and build a green cloud service system. In this paper, a cloud workflow process model, resource model, which is oriented to energy consumption, is established. The load based energy consumption calculation method and the cloud workflow scheduling optimization algorithm for energy consumption are proposed in this paper. The algorithm used in this paper is the cloud workflow grouping scheduling algorithm based on genetic algorithm (GWSGA), which is more concerned with time than the existing ones. The scheduling optimization method of intermission cost, In order to ensure the effectiveness of individual coding, the coding method of GWSGA algorithm is to group the workflow, determine the group of each task in the workflow, and always put the small task group ahead of the large group in order to ensure the effectiveness of individual coding. Then, according to the group of workflow, the task requirements and the availability of resources can get the initial population, which can effectively prevent the infeasible solution, so the biggest advantage of the GWSGA algorithm is that all the solutions of the initial population are feasible. The search time of this algorithm is very short and the efficiency is very high. It is widely used in various scientific workflows. Simulation experiments show that the GWSGA algorithm used in this paper can not only ensure the time efficiency of cloud workflow execution. And it can effectively reduce the energy consumption of the host processing task.
【学位授予单位】：浙江工商大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP308;TP18

【相似文献】