Deep learning supported database systems (part 1)

Abstract

In recent years, cardinality estimation in query optimization has been a popular area of research. With better estimation techniques, query optimizers can produce more efficient query plans that are able to directly impact the performance of Database Management Systems (DBMS). In this project, we will be extracting features from the text-based query plans of 5000 random queries on the IMDB dataset. Each of these query plans were generated by PostgreSQL, a widely used DBMS. Subsequently, we will map the extracted features to an RGB image format that can be fed as inputs into a shallow convolutional neural network (CNN). The model is tasked with a regression problem that aims to predict the actual execution time or rows returned by each query. Finally, the outputs of the model will be compared with the outputs generated by the PostgreSQL estimator.