A Survey of Database Knob Tuning with Machine Learning

As the scale of modern database management systems (DBMSs) becomes larger and larger, the configuration knobs have reached hundreds, which makes the traditional knob tuning methods can’t meet the demand. It is urgent to use the latest technology to study more intelligent and perfect knob tuning technology to ensure the high-performance operation of DBMS. In this paper, we first analyze the limitations of the traditional knob tuning methods. After that, we discuss the research status and challenges of Bayesian Optimization (BO), Reinforcement Learning (RL) and Deep Learning (DL) applied to database knob tuning problems.