An Environment-aware Anomaly Detection Framework of Cloud Platform for Improving Its Dependability

Virtualization technology is a core technology in Cloud Platform, which allows the hardware, the operating systems, and the applications running atop to be encapsulated into virtual machines (VMs). Along with the increasing scale and complexity of Cloud Platform, various faults cause the frequent downtime accidents of VMs, which has seriously lowered the dependability of Cloud Platform. Anomaly detection can detect anomalous status of VMs, while subsequent fault diagnosis can further discriminate the reasons of the detected anomalies. The former means is the foundation of the latter one. VMs are isolated one another in Cloud Platform. An anomalous VM usually does not affect other VMs. Aiming at detecting anomalous VMs in Cloud Platform, this paper proposes an environment-aware anomaly detection framework. 53 performance metrics of each VM are collected to characterize its current status. A series of processing steps are then conducted to judge whether the VMs in Cloud Platform are normal or abnormal. The experimental results show that the proposed framework can detect anomalous VMs in real time and with high accuracy rate, thus improving the dependability of Cloud Platform.