A Session-based Approach to Autonomous Database Tuning
Krisztián Mózsi, Attila Kiss · Acta Polytechnica Hungarica · 2020
By using autonomous tuning tools to optimize database systems, a lot of timeconsuming, manual work can be automated.However, self-tuning database systems are trying to optimize global metrics of efficiency, they may set back rare, but critical functions of applications that use the database.The priority of application functions cannot be expressed in existing solutions, therefore, another approach may be needed.In this paper, a session-based method is presented, where application functions are represented as sessions, by building and using language models based on previous observations.With this technique, a similarity measure can also be defined, to interpret minor differences between sessions caused by program logic, as similarity.If usage patterns appear on user level as well, it is reasonable to construct user groups along similar behavior, to utilize such patterns.As the most significant part of an autonomous solution is forecasting, a method is also presented to predict future workload characteristics, by identified user groups.Then, this approach has been evaluated in practice, mainly to determine the optimal corpus size and validate session recognition.