MAG: A performance evaluation framework for database systems

A database system includes a set of different hardware and software resources with a large number of configuration parameters that affect and control the performance of database systems. Tuning these parameters within their diverse and complex environments requires a lot of expertise and it is a time-consuming, and often a misdirected process. Furthermore, tuning attempts often lack a methodology that has a holistic view of the database system. Therefore, in this paper, we introduce MAG, a layer-by-layer tuning framework that can be used to monitor, analyze, predict, and control database configuration parameters. In particular, the framework comprises three main components: NNMonitor predicts the system performance based on its given parameters; Analyzer identifies and determines the source of the problem and then directs to deal with the problem; and NNGenerator generates the experimental training sets exploited in training NNMonitor. The proposed approach focuses on the root causes of database performance problems. The approach further strives avoiding the repetitive trial-and-error process that is a characteristic of a lot of performance-tuning efforts. We experimentally demonstrate the effectiveness of the proposed framework through an extensive set of evaluations.