A Fine-Grained Fault Detection Technique Based on the Virtual Machine Monitor

More and more applications are providing services in many important aspects of our society, any faults of these applications may lead to enormous losses. Timely detection of faults could reduce the accidental loss and make contributions to fast recovery of systems. There exist many kinds of fault detection techniques, some of them focus on the discovery of fault events but ignore the details of faults, and the others pay attention to find the exact fault in order to recover it. The former can be implemented easily but helpless to accurate fault analysis, while the accurate fault detections need deep analysis of the application in some way like white-box testing, which takes a lot of work and can hardly transplant to other applications. Besides, the source codes of the application must be available, which is not always realistic in production. In the field of cloud computing, lots of applications are running in virtual machines, and virtualization technology provides good conditions to monitor the applications. Therefore, we can collect the external characteristics of applications via virtualization technology and analyze the model between external characteristics and application faults instead of analyzing the internal logic of application directly. In this paper, we attempt to take the advantages of virtualization technology to monitor the system calls of the target application, and we don´t take insight into the target application at all. Afterwards, we establish fault detection strategy by analyzing the monitoring data and certain faults. This approach is much easier to implement and deploy than the traditional ways, and the fault coverage depends on the size of target fault set. We implement our fine-grained fault detection system based on KVM, and we conduct a series of experiments to verify the effectiveness of it. The experiments results show our detection system achieve rapid and accurate fault detection of target faults.