MORTAL: A Tool of Automatically Designing Relational Storage Schemas for Multi-model Data through Reinforcement Learning

TL;DR

MORTAL is a reinforcement learning-based tool that automatically designs relational schemas optimized for multi-model data and query performance, streamlining schema creation for complex data types.

Contribution

This work introduces MORTAL, the first tool leveraging reinforcement learning to automatically generate relational schemas for multi-model data, enhancing efficiency and performance.

Findings

Successfully generates optimized relational schemas for multi-model data

Improves query performance through schema design adjustments

Provides an interactive interface for schema and performance analysis

Abstract

Considering relational databases having powerful capabilities in handling security, user authentication, query optimization, etc., several commercial and academic frameworks reuse relational databases to store and query semi-structured data (e.g., XML, JSON) or graph data (e.g., RDF, property graph). However, these works concentrate on managing one of the above data models with RDBMSs. That is, it does not exploit the underlying tools to automatically generate the relational schema for storing multi-model data. In this demonstration, we present a novel reinforcement learning-based tool called MORTAL. Specifically, given multi-model data containing different data models and a set of queries, it could automatically design a relational schema to store these data while having a great query performance. To demonstrate it clearly, we are centered around the following modules: generating initial state based on loaded multi-model data, influencing learning process by setting parameters, controlling generated relational schema through providing semantic constraints, improving the query performance of relational schema by specifying queries, and a highly interactive interface for showing query performance and storage consumption when users adjust the generated relational schema.