Learning-based workload resource optimization for database management systems

Abstract

A DBMS training subsystem trains a DBMS workload-manager model with training data identifying resources used to execute previous DBMS data-access requests. The subsystem integrates each request's high-level features and compile-time operations into a vector and clusters similar vectors into templates. The requests are divided into workloads each represented by a training histogram that describes the distribution of templates associated with the workload and identifies the total amounts and types of resources consumed when executing the entire workload. The resulting knowledge is used to train the model to predict production resource requirements by: i) organizing production queries into candidate workloads; ii) deriving for each candidate a histogram similar in form and function to the training histograms; iii) using the newly derived histograms to predict each candidate's resource requirements; iv) selecting the candidate with the greatest resource requirements capable of being satisfied with available resources; and v) executing the selected workload.