虚拟化环境下任务调度策略研究

发布时间：2018-06-18 02:25

本文选题：虚拟化 + 任务调度　；参考：《山东大学》2014年硕士论文

【摘要】：虚拟化技术在当前数据中心中应用越来越普遍。虚拟化技术通过在同一个物理机上运行多个虚拟机来提高硬件资源的利用率。大规模的数据中心应用虚拟化技术实现资源的高效利用,可扩展性和高可用性。与传统数据中心不同,基于虚拟化的数据中心具有动态性、开放性和异构性等特点,并且以虚拟机的形式提供资源。此外,当前数据中心具有庞大的用户群体,几乎时刻都在处理海量的任务,如何合理的分配资源,高效的调度任务,使用户提交的任务处理时间较短、执行耗费较小并且使系统负载维持在一个相对均衡的状态是虚拟化环境下资源管理的重点和难点。其中,根据虚拟化环境实际状态实现合理的任务调度,是缩短任务调度长度、提高可信度、实现节能降耗等虚拟资源管理目标的关键技术。本文主要研究虚拟化环境下的任务调度策略。在总结前人工作的基础上,本文所做的研究工作包括以下几点： 1、本文根据虚拟化技术特点,深入分析虚拟化环境特征,运用图等建模技术建立虚拟化特征参数模型。通过仔细分析虚拟化环境特征,建立了虚拟化计算系统模型、虚拟机资源调度模型和任务(独立任务,非独立任务,实时任务)模型。 2、基于建立的虚拟化特征参数,针对实时(有完成时间约束)独立类型的任务提出了一种列表调度算法ELS。通过命题1证得：任务所在的虚拟机处理速度(MIPS)越慢,任务消耗的能耗越少。算法ELS基于命题1,综合考虑虚拟化环境下实时独立类型任务的能耗和调度长度,在满足任务完成时间约束的条件下尽量将任务分配到速度慢的虚拟机上,从而尽可能降低能耗。实验表明ELS能在给定的运行时间约束下显著的降低能耗。 3、针对虚拟化环境下非实时(没有完成时间约束)独立类型的任务建立了调度长度与可信度的综合权值函数,并据此提出一种混合遗传算法。该算法首先运用Min-min算法产生初始解决方案,然后运行遗传算法,并以建立的调度长度与可信度的综合权值函数为优化目标,实现可信的、同时又兼顾调度长度的任务调度方案。仿真实验证明混合遗传算法与现有较好的任务调度算法相比,能够获得良好的调度长度与可信度的总效益值。
[Abstract]:Virtualization technology is becoming more and more popular in current data centers. Virtualization improves the utilization of hardware resources by running multiple virtual machines on the same physical machine. Large-scale data center application virtualization technology to achieve efficient use of resources, scalability and high availability. Unlike traditional data centers, virtualized data centers are dynamic, open and heterogeneous, and provide resources in the form of virtual machines. In addition, the current data center has a large user group, almost all the time in dealing with a large number of tasks, how to allocate resources reasonably, efficient scheduling tasks, so that the task submitted by the user processing time is short. The key and difficulty of resource management in virtualized environment is to keep the system load in a relatively balanced state. According to the actual state of virtualization environment, the realization of reasonable task scheduling is the key technology to shorten the length of task scheduling, improve credibility, and achieve the goal of virtual resource management, such as energy saving and consumption reduction. This paper mainly studies the task scheduling strategy in virtualization environment. On the basis of summarizing the previous work, the research work in this paper includes the following points: 1. According to the characteristics of virtualization technology, this paper deeply analyzes the characteristics of virtualization environment. The virtualization characteristic parameter model is established by using graph and other modeling techniques. By analyzing the characteristics of virtualization environment carefully, the virtual computing system model, virtual machine resource scheduling model and task (independent task, non-independent task, real-time task) model are established. A list scheduling algorithm, ELS, is proposed for real-time (completion time constrained) independent tasks. It is proved by proposition 1 that the slower the processing speed of the virtual machine in which the task is located, the less energy consumption the task consumes. Based on proposition 1, the ELS algorithm considers the energy consumption and scheduling length of real-time independent task in virtualization environment, and assigns the task to a slow virtual machine under the condition of satisfying the task completion time constraints, so as to reduce the energy consumption as much as possible. Experiments show that ELS can significantly reduce energy consumption under given running time constraints. 3. A comprehensive weight function of scheduling length and reliability is established for non-real-time (not complete time constraint) independent tasks in virtualized environments. Based on this, a hybrid genetic algorithm is proposed. First, the Min-min algorithm is used to generate the initial solution, and then the genetic algorithm is run. With the established comprehensive weight function of the scheduling length and credibility as the optimization goal, a credible task scheduling scheme with both the scheduling length and the scheduling length is realized. The simulation results show that the hybrid genetic algorithm can obtain good scheduling length and total benefit value of credibility compared with the existing better task scheduling algorithm.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.01

【参考文献】