基于云用户应用的云平台虚拟机调度规划

发布时间：2018-06-08 02:45

本文选题：云应用程序 + 资源配置预估　；参考：《华北电力大学》2017年硕士论文

【摘要】：随着云计算技术的发展,IT基础设施需求已从传统的设备购买转变成按需租赁的方式。然而,人们往往会对自身程序需求资源预估不合理,导致购买的虚拟机或大或小,资源利用率不高。对云提供商来说,在资源分配前需要对用户的使用资源有一定的了解。以期更好的制定虚拟机部署策略。针对上述问题,本文提出了基于云用户应用的云平台虚拟机调度规划。分为云应用需求资源的预估、与虚拟机调度规划的设计两部分。基于云应用参数的变化,提出了一种资源配置预估方法。仿照解释器的思想,静态分析程序源码,得出任务执行的语句数;通过曲线拟合与少次数的程序运行,得出程序运行时间随参数变化的趋势。限制实验主机配置的CPU占用比例,得出参数、用户程序运行时间与CPU占用资源之间的关系。使用pascal hanoi程序对该方法进行验证,证明该方法测得参数变化下的程序运行时间与实际运行时间非常接近,误差率很小。通过对CPU占用率的限制,测得在不同的cpu-MIPS值下,程序实际运行时间与预计运行时间误差很小。提出了基于负载预测的虚拟机资源分配方案,用静态参数加动态监测相结合的双重机制避免了主机负载过重,通过预测主机负载情况,过滤掉即将达到警告值的主机;引用了二维装箱问题的BFD算法思想,对资源进行最大化利用。对本方案与轮询算法分别进行仿真实验,分析了两种算法对主机综合利用率与负载稳定性的影响;分析了本方案在警告值(A2)不同的情况下与轮询算法相比,主机综合利用率的变化情况。实验可得本方案在A2参数为0.95时,与轮询算法相比提高了资源利用率。同时主机负载变化相对平稳,主机性能相对稳定。
[Abstract]:With the development of cloud computing technology, IT infrastructure needs have changed from traditional equipment purchase to on-demand leasing. However, people often do not estimate the resources of their own programs reasonably, which leads to the purchase of large or small virtual machines and low utilization of resources. For cloud providers, it is necessary to have a certain understanding of the user's use of resources before resource allocation. In order to better formulate virtual machine deployment strategy. Aiming at the above problems, this paper proposes a cloud platform virtual machine scheduling plan based on cloud user application. It is divided into two parts: the estimation of cloud application resource and the design of virtual machine scheduling planning. Based on the variation of cloud application parameters, a resource allocation estimation method is proposed. According to the idea of interpreter, static analysis of the program source code, get the number of statements executed by the task, through curve fitting and fewer times of the program running, get the program running time with the change of parameters trend. The relationship among the parameters, the running time of the user program and the CPU occupied resources is obtained by limiting the CPU occupancy ratio of the experiment host configuration. The pascal hanoi program is used to verify the method. It is proved that the program running time measured by this method is very close to the actual running time and the error rate is very small. By limiting the CPU occupancy rate, it is found that under different cpu-MIPS values, the error between the actual running time and the estimated running time of the program is very small. A virtual machine resource allocation scheme based on load prediction is put forward. The dual mechanism of static parameter and dynamic monitoring is used to avoid the overloading of the host, and by predicting the load of the host, it filters out the host which will reach the warning value. In this paper, the idea of BFD algorithm for two-dimensional packing problem is introduced to maximize the utilization of resources. The simulation experiments of this scheme and the polling algorithm are carried out, and the effects of the two algorithms on the overall utilization ratio and load stability of the host are analyzed, and the comparison between this scheme and the polling algorithm under the different warning value of A2) is presented. The overall utilization of the host changes. Experimental results show that this scheme can improve resource utilization compared with polling algorithm when A2 parameter is 0.95. At the same time, the host load changes relatively stable, host performance is relatively stable.
【学位授予单位】：华北电力大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP302;TP393.09

【参考文献】