A Variated Monte Carlo Tree Search Algorithm for Automatic Performance Tuning to Achieve Load Scalability in InnoDB Storage Engines

Variable environmental conditions and runtime phenomena require system developers of complex business information systems to expose a set of configuration parameters to system administrators. This allows the system administrators to adjust the configuration parameters in response to the changes or in anticipation of future problems. However, these adjustments are prone to error and lack of standards due to varying levels of expertise and unpredictable short-term future states of a business information system. The purpose of this study is therefore to investigate on how to design an algorithm that proactively reconfigures bottleneck parameters without over-relying on an accurate model of an unpredictable stochastic environment. The measure of effectiveness of the reconfiguration is based on the transaction throughput and response-time latency as the server strives towards achieving load scalability. An experiment-based research design was used in conjunction with hypothesis testing to analyze the experiment results. Over 95% of the experiment results provided evidence that the use of the distributed database architecture with the designed algorithm running on each node, resulted in faster transaction throughput and slower response-time latency. The study builds on automatic performance tuning, reinforcement learning, and simulation-based inferential statistics literature by combining the three fields. A Monte Carlo Tree Search with variated selection, expansion, and simulation stages was used as the core of the designed algorithm