Automated Test Scheduling using Adaptive Hybrid Prediction Technique

In a typical data center, there is always an ongoing need to isolate faulty components. Diagnostic and Regression tests are generally tending towards automation. However, the diagnostic tools also put the underlying hardware to various levels of stress. It is challenging to select appropriate test tools and schedule them in such a way that they can help uncover maximum defects and ensure minimal disturbance to the live customer setup. In this paper, we propose a technique to automatically schedule tests on a target system with minimal disturbance to the workload using a real-time adaptive hybrid predictive model. The models we use are trained to predict resource utilization in a fast and accurate manner. This solution enhances the critical decision-making ability of an admin by scheduling the regression or diagnostic tests accurately. Schedules are recommended based on actual resource utilization and are spaced out at intervals when low resource utilization is predicted. This ensures minimal downtime for maintenance and helps meet customer SLA. We also propose a technique to automate the selection process of the best fit time-series model based on analysis of data, which in due course would reduce the prediction overhead by half. This solution can work alongside any existing management framework or can be designed as a standalone tool.