Research And Implementation Of Virtual Machine Monitoring In The Cloud Computing Environment

Today, cloud computing has become the most widely used Internet service model. Monitoring the resource in cloud platform need to be effective, timely, efficient, low-overhead, which is essential to ensure the quality of service of cloud computing. Resource monitoring is also the basis of next system management, load management and balancing work. Virtual machine monitor in cloud platform conclude obtaining the operating status information of the virtual machine, analysis of virtual machine anomaly and fault prediction.This paper puts forward an anomaly detection algorithm based on clustering. On this basis, a virtual machine anomaly detection system in a cloud environment is designed and implemented. The main function of the system is to monitor the state of the virtual machine nodes, which run on the cloud platform, and warn the fault.Anomaly detection algorithm based on clustering is divided into two parts, which are modeling method based on clustering and anomaly analysis method based on the nonparametric CUSUM. In the first part, K-means and K-modes clustering are used for modeling. First, we input training data and specify the cluster center. Then virtual machine state model with two methods, and we obtain and make amendments to the results. Finally, according to the modeling results, we divide virtual machine state into three categories, namely, normal, abnormal, fault. Compared two methods, for numeric data type in this article k-means performs better. The second part deals with the data divided into abnormal categories. When the system detects an abnormal state of the virtual machine, we use CUSUM algorithm, increasing the frequency of sampling. When accumulation of abnormal data reaches threshold, a forecast alert will alarm.Virtual machine monitoring System in cloud computing environments implemented on Hadoop and Spark cloud computing platform. In the system with centralized monitoring architecture, master and slave node from the virtual machine is designed. The main function of slaves is to collect data of running state of the virtual machine's, to send data to the masters by Kafka message system, and stored data in Rsdis database. The master receives the detection data through the message system, and using the relevant algorithm for anomaly analysis and fault warning. The master also needs to have an interface for the user to check the running state of the virtual machine and alarm information. Experimental result show that the monitoring system under the Spark platforms can achieve the desired functionality, while result on Hadoop is poor at timeliness.