Abstract:We propose a new approach of NoSQL database index selection. For different workloads, we select different indexes and their different parameters to optimize the database performance. The approach builds a deep reinforcement learning model to select an optimal index for a given fixed workload and adapts to a changing workload. Experimental results show that, Deep Reinforcement Learning Index Selection Approach (DRLISA) has improved performance to varying degrees according to traditional single index structures.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to dynamically select the optimal index structure and its parameters according to different workloads in NoSQL databases in order to optimize database performance. Specifically, the traditional single - index structure is difficult to handle workloads in many different situations, and the existing NoSQL database solutions usually do not combine several different index structures for simultaneous use, which is a common requirement in practical applications. In addition, index selection needs to adapt to most workloads and be as compatible as possible with more index types, and as the database is used, the workload will change, and the optimal index selection in different hardware environments also varies greatly. To solve these problems, the author proposes a NoSQL database index selection method based on deep reinforcement learning (DRLISA). This method can recommend index structures and their parameters, uses a reinforcement learning model as a heuristic method for index selection under dynamic workloads, and selects the optimal index based on the Q - value in the reinforcement learning network. Through this method, DRLISA can not only reduce the cost for system administrators to select appropriate indexes, but also provides a good framework for developing an automatic index selection system. The main contributions of the paper include: 1. Recommending index structures while recommending index parameters. 2. Using a reinforcement learning model to handle index selection under dynamic workloads. 3. Selecting indexes based on the Q - value in the reinforcement learning network. The experimental results show that DRLISA has different degrees of performance improvement compared with the traditional single - index structure under different workloads, which proves the effectiveness and adaptability of this method.

Index Selection for NoSQL Database with Deep Reinforcement Learning

ACDRL: A Constraint-guided Dynamic Reinforcement Learning for Index Selection

Dynamic Index Construction with Deep Reinforcement Learning

IA2: Leveraging Instance-Aware Index Advisor with Reinforcement Learning for Diverse Workloads

Automated Database Indexing using Model-free Reinforcement Learning

DITune: A Reinforcement Learning Based Framework for Automated Database Index Selection

Progressive Neural Index Search for Database System

Learning to Optimize Join Queries With Deep Reinforcement Learning

A Survey on Deep Reinforcement Learning for Data Processing and Analytics

A Scalable Learned Index Scheme in Storage Systems

ASLM: Adaptive Single Layer Model for Learned Index

Solving Inventory Management Problems Through Deep Reinforcement Learning

GSBRL : Efficient RDF Graph Storage Based on Reinforcement Learning.

Intrinsically Motivated Reinforcement Learning Based Recommendation with Counterfactual Data Augmentation

Deep Reinforcement Learning for Large-Scale Inventory Management

NoSQL Database Tuning through Machine Learning

Top-aware reinforcement learning based recommendation

Selective Data Collection Method for Deep Reinforcement Learning

Online Control of Adaptive Large Neighborhood Search using Deep Reinforcement Learning

Reinforcement Learning for Node Selection in Mixed Integer Programming

NeurDB: On the Design and Implementation of an AI-powered Autonomous Database