云资源管理中预测方法的研究与实现

发布时间：2018-03-09 01:35

本文选题：云计算　切入点：虚拟资源管理　出处：《复旦大学》2013年硕士论文　论文类型：学位论文

【摘要】：随着云计算技术的不断发展,虚拟资源管理已成为其中关键技术之一。为了提高资源利用率,获得更大的收益,云计算需要对资源进行有效的管理。通过对一些虚拟资源管理方法的研究我们发现这些方法均采用了预测技术。应用的负载通常是动态变化的,因此其所需的资源也在不断变化。由于资源从分配到使用存在一定的时延,云系统需要提前为应用准备好所需资源。为了解决这个问题,预测技术被引入到了云资源管理方法中。通过对应用负载特征分析我们发现负载存在一定的模式：周期模式,非周期模式或混合模式。目前还没有一种预测模型能够适用于所有模式的负载。为了更准确的预测应用负载,本文设计了模式感知预测方法。该预测方法能够识别应负载的模式,并选择相应的预测模型进行预测,在负载模式发生变化后及时感知这一变化,同时调整预测模型以适应负载的新模式。在预测中,误差是不可避免的,为了减少预测误差,我们采用人工神经网路(ANN)和马尔科夫链方法对预测值进行修正。为了验证预测模型的准确性和整个预测方法的自适应性,我们采用虚拟机的CPU利用率作为实验数据进行测试。实验结果表明我们的预测模型能够较准确地预测应用的负载；预测方法能够自适应的感知模式变化并对预测模型做出调整；通过误差修正,我们的预测方法的准确性得到了进一步提高。
[Abstract]:With the development of cloud computing technology, virtual resource management has become one of the key technologies. Cloud computing requires efficient resource management. Through the study of some virtual resource management methods, we find that these methods use predictive techniques. The load of applications is usually dynamic. As a result, the resources required are also constantly changing. Because there is a delay in resource allocation and usage, cloud systems need to prepare the required resources for the application in advance. To solve this problem, Prediction technology is introduced into cloud resource management. By analyzing the characteristics of application load, we find that there is a certain pattern of load: periodic mode, Aperiodic mode or mixed mode. There is no prediction model that can be applied to all modes. In order to predict the application load more accurately, In this paper, a pattern perception prediction method is designed, which can identify the load pattern and select the corresponding prediction model to predict the change of the load mode. In order to reduce the prediction error, the artificial neural network (Ann) and Markov chain method are used to modify the prediction value. In order to verify the accuracy of the prediction model and the self-adaptability of the whole prediction method, we use the CPU utilization of the virtual machine as the experimental data to test. The experimental results show that our prediction model can accurately predict the load of the application. The prediction method can adaptively perceive the pattern change and adjust the prediction model, and the accuracy of our prediction method is further improved by error correction.
【学位授予单位】：复旦大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP3

【共引文献】