Research On System Behavioral Anomaly Automatic Detection In Cloud Computing Environment

As the information technology develops fast nowadays, modern computer systems become increasingly complex in both infrastructure and usage. Especially the cloud computing which has become more and more popular to be an important and commonly used form of computation service because of the development of high speed network and virtualization technology. Cloud computing systems normally have much complex internal structure and design to support various kinds of important service and application requirements. It has been critical to ensure the functionality and healthiness of a cloud computing system and auto anomaly detection, alert or even prediction is the basis for that. Although there have been some research result of point anomaly detection techniques, they are still not enough when actual enterprise cloud platform is concerned in both precision and scalability. Furthermore, behavior anomaly detection still lacks effective approaches because of its inherent complexities. This thesis will present algorithms which combine techniques of machine learning, multivariate analysis, time series analysis and random process, etc. mainly targeting behavioral anomaly detection. An online detection system based on above algorithms is also introduced which can provide online anomaly detection and results visualization. Intense and comprehensive experiments on various data sets have been make. Results show that these algorithms have great performance under complex real situations and outperform traditional techniques.