基于资源监控的Linux电源管理系统的设计与实现

发布时间：2018-09-03 20:02

【摘要】：随着计算机的普及，计算机节能已经成为建设节约型社会的重要部分，也是关键的用户需求。电源管理是计算机节能的有效方法之一，本文中我们重点研究其中的一个重要方向：动态频率调节技术。动态频率调节技术已经被证实是一个获得低功耗的高效方法。动态频率调节技术的本质是动态调节CPU的频率，以便提供刚刚好的电路速率来处理系统负载以满足总的计算时间和总的约束，从而降低能耗。首先，本文介绍了低功耗技术、电源管理相关技术、动态频率调节技术，并介绍了4种经典的动态频率调节算法: PAST、AVGn、FLAT、PEAK算法。本文研究了针对XEN虚拟机的资源监控技术，提出了适合本系统的资源监控方法：采用Xentrace、Xenstore、虚拟机中读取数据三种方式来获取CPU、内存、IO相关信息。另外，本文研究了回归分析等预测算法并提出了适用于本系统的预测算法。其次，基于提出的预测算法，，本文最终设计并实现了基于资源监控的Linux电源管理系统。本系统包括三个模块：资源监控模块、预测分析模块、频率调节模块。资源监控模块介绍了模块功能的设计和信息采集的设计。预测分析模块介绍了本模块的设计与实现。频率调节模块利用Linux的cpufreq机制来实现，介绍了cpufreq系统、内部机制以及设置CPU频率的方式。最后，对系统进行了测试。分为验证预测分析模块和测试节能效果两部分。实验结果表明本文提出的预测算法可以很好地拟合物理机、虚拟机的CPU、内存、IO利用率，本文实现的电源管理系统能有效地节能。
[Abstract]:With the popularity of computers, computer energy conservation has become an important part of building a conservation-oriented society, and also a key user demand. Power management is one of the effective methods of computer energy saving. In this paper, we focus on one of the important directions: dynamic frequency regulation technology. Dynamic frequency regulation has been proved to be an efficient way to achieve low power consumption. The essence of the dynamic frequency regulation technique is to dynamically adjust the frequency of the CPU so as to provide a good circuit rate to process the system load to meet the total computing time and constraints, thus reducing the energy consumption. Firstly, this paper introduces the low-power technology, power management technology, dynamic frequency regulation technology, and introduces four classical dynamic frequency regulation algorithm: PAST,AVGn,FLAT,PEAK algorithm. In this paper, the resource monitoring technology of XEN virtual machine is studied, and the method of resource monitoring suitable for this system is put forward. Three ways of reading data from Xentrace,Xenstore, virtual machine are used to obtain the information about CPU, memory IO. In addition, the prediction algorithms such as regression analysis are studied and a prediction algorithm suitable for this system is proposed. Secondly, based on the proposed prediction algorithm, this paper finally designs and implements the Linux power management system based on resource monitoring. The system consists of three modules: resource monitoring module, prediction and analysis module, frequency regulation module. The resource monitoring module introduces the design of module function and information collection. Predictive analysis module introduces the design and implementation of this module. The frequency regulation module is implemented by the cpufreq mechanism of Linux. The cpufreq system, the internal mechanism and the way of setting the CPU frequency are introduced. Finally, the system is tested. It is divided into two parts: verify prediction analysis module and test energy saving effect. The experimental results show that the prediction algorithm proposed in this paper can fit well the utilization ratio of CPU, memory IO of virtual machine and the power management system realized in this paper can save energy effectively.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP303.3

【相似文献】