WNN-based Fast Event Pattern Detection and Prediction Using Reversed Pattern Tree for Cloud System Reliability Management

Cloud service systems need an efficient and flexible platform to offer services and computational resources to users. However, emergent event detection and prediction for platform reliability management is still an unsolved problem for most cloud service providers, especially for virtualized platforms. Hardware failure, software errors, outside attacks, and mis-actions of virtual machines make cloud platforms unstable and unreliable. In order to avoid critical events affecting reliability, resources, applications, and services can be rescheduled to get around predicted failures and mitigate potential impacts. In a cloud platform, different layers of service execution may generate different types of system events. Events from different layers can affect a system's stability together. Thus, event pattern detection and prediction are especially important for reliability and stability management. In this paper, we propose a framework to detect and predict system critical event patterns from different service types and execution layers using a reversed pattern tree augmented mechanism, so that detection and prediction can speed up. Meanwhile, wavelet neural network is also used to control the balance between prediction accuracy and speed, so that system reliability and stability can be improved. The effectiveness of this framework is tested and confirmed in a real cloud environment.