云平台的资源监控与弹性伸缩技术研究与实现

发布时间：2018-07-08 09:53

本文选题：云监控系统 + 推拉混合模式　；参考：《北京邮电大学》2017年硕士论文

【摘要】：云计算的快速发展满足了大规模计算、海量数据存储的要求。对于云平台来说资源繁多、情况复杂,有必要利用监控系统监测整个平台运行状况、资源使用情况。为了实现云计算资源的高效利用,弹性伸缩服务也成为不可或缺的一部分。良好的监控系统可以方便用户对云资源的合理调度,也为发现故障原因提供数据支持。本文针对目前云市场已经实现的监控系统不足之处进行了研究,设计了 WEB界面、两级数据库存储、及时报警多位一体的高内聚、低耦合的架构。在提高监控系统效率方面,本文从数据的产生、传送出发分别提出利用/proc文件系统进行数据采集、利用推拉混合模式进行数据传输。在提高监控系统可靠方面,本文设计和实现了故障监测、日志分析、通知报警组成的故障处理子系统,并引用一主多从的高可用数据存储模型进行数据持久化保障。当前云市场的弹性伸缩服务在进行资源再分配的时候大都需要强制暂停用户的应用任务,给用户带来比较差的体验感。针对上述问题,本文提出基于负载预测的动态弹性伸缩模型,在不影响用户使用的情况下提高了资源利用率、降低了云平台的资源投入成本。这个模型分为两部分,第一部分是负载预测,根据负载历史数据的相似性和周期性,运用线性回归和负载数字串匹配两种性能互补的算法预测下一时刻负载需求量;第二部分是弹性伸缩,在满足资源伸缩需求的前提,利用整数规划的算法求解出成本最小的水平伸缩量和垂直伸缩量,并进行相应的资源扩展和释放。为了验证监控系统的高效性、可靠性,本文对设计方案进行了实现和应用,对推拉混合模式进行了实验和分析。为了证明基于负载预测弹性伸缩模型的优越性,本文做了相应的对比实验,表明本文负载预测的平均误差为8%,比自回归滑动平均模型指数平滑模型都低,本文伸缩方法比水平伸缩方法节省成本5.88%,降低了资源使用成本。
[Abstract]:The rapid development of cloud computing meets the requirements of large-scale computing and massive data storage. For the cloud platform, there are many resources and complex situations, so it is necessary to monitor the whole platform running condition and resource usage by using the monitoring system. In order to achieve the efficient use of cloud computing resources, flexible expansion services have become an indispensable part. A good monitoring system can facilitate users to reasonably schedule cloud resources and provide data support for fault detection. In this paper, the shortcomings of the monitoring system which has been realized in the cloud market have been studied, and a Web interface, two-level database storage, high cohesion and low coupling architecture with timely warning and multi-bit integration have been designed. In order to improve the efficiency of the monitoring system, this paper proposes to collect the data by using the Proc file system and to transmit the data using the push-pull mixed mode from the point of view of the data generation and transmission. In order to improve the reliability of the monitoring system, this paper designs and implements a fault processing subsystem composed of fault monitoring, log analysis, notification and alarm, and uses a high availability data storage model of master, multi-slave to ensure data persistence. At present, the flexible expansion service in cloud market needs to force users to suspend their application tasks when redistributing resources, which brings users a relatively poor sense of experience. To solve the above problems, a dynamic elastic expansion model based on load prediction is proposed in this paper, which improves the resource utilization and reduces the cost of resource input of cloud platform without affecting the user's use. The model is divided into two parts. The first part is load forecasting. According to the similarity and periodicity of load historical data, two complementary algorithms of linear regression and load digital string matching are used to predict the load demand at the next moment. The second part is elastic expansion. In order to meet the requirement of resource expansion, we use integer programming algorithm to solve the minimum cost of horizontal expansion and vertical expansion, and carry out the corresponding resource expansion and release. In order to verify the high efficiency and reliability of the monitoring system, the design scheme is implemented and applied, and the push-pull mixed mode is tested and analyzed. In order to prove the superiority of elastic expansion model based on load prediction, a comparative experiment is made in this paper, which shows that the average error of load prediction in this paper is 8, which is lower than the exponential smoothing model of autoregressive sliding average model. In this paper, the expansion method than horizontal expansion method cost savings 5.88, reducing the cost of the use of resources.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.09

【相似文献】