云计算服务平台资源管理系统

发布时间：2018-09-07 07:12

【摘要】：作为一种新兴的技术,云计算将IT基础设施、平台和软件以服务的形式通过网络以按需分配的方式对外提供。云计算通过虚拟化的方式将计算、存储、网络等资源构建成统一的资源池,有效降低资源分配和管理的难度。如何高效、合理地管理资源池,实现云计算的动态性、可伸缩性等特点显得尤为重要。目前主流的云计算服务平台如OpenStack、CloudStack等依赖人工配置方式实现资源管理。由于业务的资源需求规模和系统负载呈现周期性波动,用户无法及时准确地调整资源,导致出现业务运行异常和资源浪费等问题。针对这些问题,本论文提出了一种基于OpenStack的资源自适应管理系统。该系统基于平台资源的监控数据进行智能分析和决策,动态调整虚拟机资源。本论文的主要研究工作分成了四个部分：首先,设计了一个采用DSA (Disco very-Strategy-Action)思想的资源自适应管理框架。该框架包括资源发现、资源调度决策和资源调整实施三个部分。资源发现部分负责监测资源池的资源状况和接收用户资源请求信息。资源调度决策部分基于上述信息生成资源调整的操作指令。资源调整实施部分负责执行操作指令。通过这三部分的协同工作,提高了系统响应资源需求的实时性。然后,设计实现了基于C/S架构的资源监控子系统。该子系统包括客户端和服务端两个模块。客户端模块完成对物理主机和虚拟机的CPU利用率、内存等性能数据的周期性采集。服务端模块负责汇总客户端采集的监控数据,完成数据持久化存储,并基于可配置的阂值检测资源负载触发告警。接着,设计实现了资源管理决策子系统。该子系统负责处理资源负载告警和创建虚拟机请求。对于虚拟机资源告警,系统自动统计历史监控数据,生成合理的资源调整决策,借助OpenStack提供的虚拟机配置调整和虚拟机迁移的接口完成资源的动态调整。这样有效减少了瞬时峰值引起的不必要调整。对于创建虚拟机的请求,采用贪心算法选择合适的主机来放置新建虚拟机。该子系统在满足自适应业务资源需求的同时保证平台资源利用最大化。最后,本论文将资源监控和管理决策两个子系统的功能封装成标准REST API。基于这些API和OpenStack服务接口,设计实现了云计算服务平台的Web控制台。该控制台支持用户定制虚拟机的资源监控和自动调整功能,可视化呈现虚拟机的监控数据等。本论文对实现的云计算服务平台资源管理系统进行了实验和测试,并将实验室的多媒体会议系统运行到平台上以验证资源管理系统的有效性。实验结果表明,所提的资源管理系统能及时有效地适应业务资源需求和系统负载的变化,实现了预期的功能需求。
[Abstract]:As an emerging technology, cloud computing provides IT infrastructure, platforms and software as services through the network and distributes them on demand. Cloud computing constructs computing, storage, network and other resources into a unified resource pool through virtualization, which effectively reduces the difficulty of resource allocation and management. It is very important to manage resource pool efficiently and reasonably to realize the dynamic and scalability of cloud computing. Current mainstream cloud computing service platforms such as OpenStack,CloudStack rely on manual configuration to achieve resource management. Due to the periodic fluctuation of resource demand scale and system load, users can not adjust resources accurately and timely, which leads to problems such as abnormal operation of business and waste of resources. To solve these problems, this paper proposes a resource adaptive management system based on OpenStack. The system makes intelligent analysis and decision based on monitoring data of platform resources and dynamically adjusts virtual machine resources. The main work of this thesis is divided into four parts: firstly, a resource adaptive management framework based on DSA (Disco very-Strategy-Action is designed. The framework includes three parts: resource discovery, resource scheduling decision and resource adjustment implementation. The resource discovery section is responsible for monitoring the resource status of the resource pool and receiving user resource request information. The resource scheduling decision part generates the operation instruction of resource adjustment based on the above information. The Resource Adjustment implementation is responsible for executing operational instructions. Through the collaborative work of the three parts, the real-time response of the system is improved. Then, the resource monitoring subsystem based on C / S architecture is designed and implemented. The subsystem includes two modules: client and server. The client module completes the periodic collection of CPU utilization, memory and other performance data of the physical host and virtual machine. The server module is responsible for collecting the monitoring data collected by the client, completing the data persistence storage, and detecting the resource load trigger alarm based on the configurable threshold value. Then, the resource management decision subsystem is designed and implemented. The subsystem is responsible for handling resource load alarms and creating virtual machine requests. For the virtual machine resource alarm, the system automatically statistics the historical monitoring data, generates reasonable resource adjustment decision, and completes the dynamic resource adjustment with the help of the virtual machine configuration adjustment and the virtual machine migration interface provided by OpenStack. This effectively reduces the unnecessary adjustment caused by the instantaneous peak. For the request of creating virtual machine, the greedy algorithm is used to select the appropriate host to place the new virtual machine. The subsystem not only meets the needs of adaptive business resources, but also ensures the maximum utilization of platform resources. Finally, the functions of resource monitoring and management decision subsystem are encapsulated into standard REST API.. Based on these API and OpenStack service interfaces, the Web console of cloud computing service platform is designed and implemented. The console supports resource monitoring and automatic adjustment of custom virtual machines, and visualizes monitoring data of virtual machines. In this paper, the resource management system of the cloud computing service platform is tested and tested, and the multimedia conference system of the laboratory is run on the platform to verify the effectiveness of the resource management system. The experimental results show that the proposed resource management system can effectively adapt to the changes of business resource requirements and system load and achieve the expected functional requirements.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.07

【参考文献】