Design And Implementation Of An Anomaly Detection System For Cloud Platforms Based On Multi-attribute Information

Today, Cloud computing has become a mainstream model of the global IT industry and the development of various fields can’t leave the Cloud computing. In order to improve the quality of service in the Cloud computing platform, we need comprehensive monitoring on a cloud platform. Due to Cloud computing resources has the characteristic such as diversity, complexity, and brings great challenge to the monitoring task. Cloud monitoring of the key problems are mainly involved in monitoring information acquisition, monitoring data anomaly judgment, abnormal monitoring data information transmission and abnormal warning.This paper puts forward a kind of anomaly detection methods that based on Principal component Analysis (Principal Component Analysis, PCA) and multiple attribute information of the virtual machine. On this basis, designed and implemented a cloud platform for anomaly detection system based on multiple attribute information, the monitoring system can carry on the resources to the Hadoop cluster nodes in the virtual machine monitor and fault early warning.Anomaly detection method research based on PCA and multiple attribute information including PCA algorithm as the core of multiple attribute information classification and non-parametric CUSUM algorithm as the core of anomaly detection. The former first by PCA algorithm to multiple attributes into a new set of data independent of each other comprehensive data instead of the original multidimensional data, and then through the analysis and calculation to the integrated data is divided into "normal", "exceptionally" and "fault".The latter mainly aimed at the abnormal data, when detect abnormal data, the system will start the CUSUM detection mechanism, speed up the sampling frequency, when testing sequence from negative to positive cumulative abnormal, when the accumulated value reaches the threshold anomaly threshold sends a warning. The experimental results show that the multiple attribute information anomaly detection algorithm based on PCA can significantly abnormal in the running state of the virtual machine before accurately detect abnormalities, and has a smaller time delay.Cloud platform anomaly detection system based on multiple attribute information design including the node function design and controller design. This experiment mainly implemented in Xen cloud platform. The realization of the function of node machine first by Sysstat system performance collection tool of node information acquisition, and then through the background daemon to sort the information and deposited in the Mysql database, at last, an anomaly detection algorithm of abnormal test and faults. Master machine is an anomaly detection system of the human-computer interaction interface, responsible for the sending and receiving of all commands, and have abnormal alarm, warning, record store and the function of each node resource utilization status view.The experimental results show that the system can achieve the expected design function.