基于线性回归与马尔科夫链相结合的云资源监控预测算法研究与实现

发布时间：2018-06-24 03:22

本文选题：监控 + 云计算　；参考：《浙江大学》2017年硕士论文

【摘要】：随着云计算的发展,其提供的功能也越来越丰富,管理的计算机集群规模也逐渐上升,云基础架构和部署的高效管理是目前的一个引人注目的话题。监控工具和监控技术在这方面可以发挥重要作用,收集相关的信息,提供给管理员或用户以便他们做出相应的决策。本文通过对自搭建的OpenStack云环境得到的实验数据以及来自某公有云平台的监控数据的分析,发现并找出数据中隐藏的规律、特征。通过分析了线性回归算法和马尔科夫链算法两个预测算法在预测云环境的监控数据的应用以及它们的缺陷,利用发现的规律、特征对预测算法进行优化改进,提高算法预测的命中率、降低算法的某些开销。本文的主要研究工作如下:1)研究云计算的监控,通过对历史数据的分析,发现数据中的规律、特征,将一天分为两个不同的时间段:忙碌期和平稳期,在不同的时期内由于用户访问量的差异而引起资源消耗的差异,进而影响到在不同时期内收集到的数据特征的不同;2)研究时间序列预测算法,针对数据的特征,提出新的算法模型——结合线性回归算法与马尔科夫链算法(line discrete time markov chain,称为L-DTMC算法),并且根据不同的时间段采用不同的算法,然后根据误差容忍度(Error Tolerant Degree,ETD)决定是否需要更新算法模型以及是否需要向服务器发送更新数据;3)通过部署一个小型的云平台,分别针对被监控节点与数据服务器的数据的一致性、算法的预测效果以及执行该算法对系统的影响这三个方面对本文提出的算法进行模拟验证实验,通过分析实验结果验证了本文工作的有效性。
[Abstract]:With the development of cloud computing, the functions provided by cloud computing are becoming more and more abundant, and the scale of managed computer clusters is increasing gradually. The efficient management of cloud infrastructure and deployment is a noticeable topic at present. Monitoring tools and monitoring techniques can play an important role in this regard, collecting relevant information and providing it to administrators or users so that they can make appropriate decisions. Based on the analysis of the experimental data obtained from OpenStack cloud environment and the monitoring data from a public cloud platform, the hidden rules and features of the data are found and found in this paper. Based on the analysis of the application of linear regression algorithm and Markov chain algorithm in the monitoring data of cloud environment prediction and their defects, the prediction algorithm is optimized and improved by using the discovered rules and features. Improve the hit ratio of the algorithm prediction, reduce some of the cost of the algorithm. The main research work of this paper is as follows: 1) Research cloud computing monitoring, through the analysis of historical data, find the laws and characteristics of the data, divide the day into two different time periods: busy period and stationary period. In different periods, the difference of resource consumption caused by the difference of user visits will affect the different features of the data collected in different periods. (2) to study the time series prediction algorithm, aiming at the characteristics of the data. A new algorithm model, combining linear regression algorithm and Markov chain algorithm called L-DTMC algorithm, is proposed, and different algorithms are used according to different time periods. Then according to error tolerance (ETD) to decide whether to update the algorithm model and whether to send update data to the server. By deploying a small cloud platform, the consistency of the data between the monitored node and the data server is analyzed, respectively. The prediction effect of the algorithm and the effect of executing the algorithm on the system are simulated and validated. The effectiveness of the proposed algorithm is verified by the analysis of the experimental results.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP301.6;O211.62

【参考文献】