Toward Anomaly Detection in IaaS Cloud Computing Platforms

In order to monitor the running status of IaaS cloud computing platforms, performance metric data are collected to perform anomaly detection for IaaS cloud computing platforms and determine whether the IaaS cloud computing platforms fail to run normally. However, it is challenging to effectively detect performance anomalies from a large amount of noisy and high dimensional performance metric data. In this paper, an efficient anomaly detection scheme is proposed for IaaS cloud computing platforms. The proposed scheme first designs a global locality preserving projection algorithm to perform feature extraction on performance metric data, and then introduces a local outlier factor algorithm to detect anomalies. A series of experiments are conducted on a private cloud computing platform. Experimental results show that our proposed global locality preserving projection algorithm outperforms the principal components analysis algorithm and the locality preserving projection algorithm and our proposed anomaly detection scheme is better than the state-of-the-art schemes for IaaS cloud computing platforms.