基于虚拟化技术的集群自适应功耗管理

发布时间：2018-08-11 09:21

【摘要】：数据中心作为云计算平台的基础依托设施,正在云计算技术的推动下,以前所未有的规模扩张,然而,数据中心的高能耗、资源利用率低、对环境的污染等问题,一直以来极大的制约着数据中心的发展。虚拟化技术的应用,为解决上述问题提供了一个很好的途径,现今大多数降低数据中心能耗的研究,都是依托在虚拟化技术之上,虚拟化的服务器集群在节能方面有很多优势,通过对计算资源的有效管理和调度,动态调节服务器的状态,可以有效减小数据中心的能耗。本课题源于国家自然基金项目,希望通过现有的相对成熟的虚拟化技术构建一种新型的数据中心功耗管理系统,在保证服务质量的前提下,通过资源整合和动态调度,来降低数据中心的整体能耗。根据这一目标,本文在开源云资源管理软件OpenNEbula的基础上,提出了CREMS云资源管理系统。本论文主要的研究和创新包括以下几点: 1)构建一种针对数据中心的新型资源管理系统,该系统将数据中心的所有物理机和虚拟机进行统一管理,系统中物理机和虚拟机的CPU利用率、内存利用率状态以及应用程序的服务质量等信息能够被及时获取,便于管理和监控虚拟化集群。 2)提高数据中心资源分配的动态性和合理性,本文提出的资源管理系统能够收集到所有虚拟机和物理机的状态信息,对整个数据中心的资源进行调节,在局部根据各个虚拟机的服务优先级和性能指标的不同,动态调节虚拟机获得相应的资源,在保证服务质量的前提下优化局部物理资源分配,从整体上通过负载整合和动态开关闭物理机,来提高资源利用率,降低整体功耗。 3)以网页或者数据库这些应用服务器中的响应时间作为描述应用性能的指标,据此结合CREMS系统监视到的动态变化的虚拟机的资源利用率,CREMS做出合理的资源需求预测,,根据预测负载来调整虚拟机的资源分配,通过虚拟机迁移、挂起等操作,实现调节资源分配和节能的目标。 4)研究调度算法和策略,根据采集到的虚拟机和物理机的数据信息,在系统中设计高效的调度算法,动态的调度每个虚拟机资源的分配,以及调整物理机的功耗状态,使得资源分配更加合理有效,保证系统运行的稳定性和持久性。本论文首先验证了CPU利用率和功耗之间的关系,结果显示服务器能源的绝大部分是被CPU消耗掉,然后在搭建的试验平台上运行基于OpenNEbula的数据中心资源管理系统CREMS,通过动态调节负载,对CREMS系统的预测和调度功能进行测试,实验结果表明,该系统能够在保证服务质量的前提下,减少大约12%的整体功耗。
[Abstract]:Data center, as the basic infrastructure of cloud computing platform, is expanding on an unprecedented scale driven by cloud computing technology. However, the data center has many problems, such as high energy consumption, low utilization of resources, pollution to the environment, etc. The development of data center has been greatly restricted. The application of virtualization technology provides a good way to solve the above problems. Nowadays, most of the research on reducing energy consumption of data center is based on virtualization technology. The virtualized server cluster has many advantages in energy saving. Through the efficient management and scheduling of computing resources and dynamically adjusting the state of the server, the energy consumption of the data center can be effectively reduced. This topic is originated from the National Natural Fund project, hoping to construct a new data center power management system through the existing relatively mature virtualization technology, under the premise of ensuring the quality of service, through the integration of resources and dynamic scheduling. To reduce the overall energy consumption of the data center. According to this goal, this paper puts forward the CREMS cloud resource management system based on the open source cloud resource management software OpenNEbula. The main research and innovations of this paper are as follows: 1) A new resource management system for data center is constructed, which manages all physical computers and virtual machines in the data center. Information such as CPU utilization, memory utilization status and application quality of service of the physical machine and virtual machine in the system can be obtained in time. It is easy to manage and monitor virtualized cluster. 2) improve the dynamic and rationality of resource allocation in data center. The resource management system proposed in this paper can collect the state information of all virtual machines and physical machines. Adjust the resources of the whole data center, dynamically adjust the virtual machine to obtain the corresponding resources according to the different service priority and performance index of each virtual machine, and optimize the local physical resource allocation under the premise of guaranteeing the quality of service. On the whole, through load integration and dynamic turn on and off the physical machine, to improve the resource utilization, reduce the overall power consumption. 3) take the response time in the application server such as web page or database as the index to describe the application performance. According to this, the resource utilization ratio of dynamic virtual machine monitored by CREMS system is combined with CREMS to make reasonable resource demand prediction. According to the forecast load, the resource allocation of virtual machine is adjusted, and the virtual machine is migrated, suspended and so on. To achieve the goals of regulating resource allocation and energy saving. 4) the scheduling algorithm and strategy are studied. According to the collected data of virtual machine and physical machine, an efficient scheduling algorithm is designed in the system. The resource allocation of each virtual machine is dynamically scheduled and the power state of the physical machine is adjusted to make the resource allocation more reasonable and effective to ensure the stability and persistence of the system. This paper first verifies the relationship between CPU utilization and power consumption. The results show that most of the server energy is consumed by CPU. Then run the data center resource management system based on OpenNEbula on the test platform, and test the prediction and scheduling function of the CREMS system by dynamically adjusting the load. The experimental results show that, The system can reduce the overall power consumption by about 12% under the premise of guaranteeing the quality of service.
【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP391.9;TP308

【共引文献】