Overcoming Training Costs in Index Advising: The Need for IA2

Table of Links

Abstract and 1. Introduction

1. Related Works

   2.1 Traditional Index Selection Approaches

   2.2 RL-based Index Selection Approaches

2. Index Selection Problem

3. Methodology

   4.1 Formulation of the DRL Problem

   4.2 Instance-Aware Deep Reinforcement Learning for Efficient Index Selection

4. System Framework of IA2

   5.1 Preprocessing Phase

   5.2 RL Training and Application Phase

5. Experiments

   6.1 Experimental Setting

   6.2 Experimental Results

   6.3 End-to-End Performance Comparison

   6.4 Key Insights

6. Conclusion and Future Work, and References

2.2 RL-based Index Selection Approaches

Recent advancements have seen the application of Reinforcement Learning (RL) to the index selection problem, offering novel approaches that promise to overcome some of the limitations of traditional methods.

\ DRLinda, introduced by Sadri et al. [10], targets cluster databases and, while innovative in its focus on such environments, does not support multi-attribute indexes and lacks a public implementation for validation against state-of-the-art methods.

\ Lan et al. [7] propose an RL-based solution capable of identifying multi-attribute indexes. Despite this advancement, their approach does not model workload representation, limiting its ability to generalize to new or unseen workloads and potentially constraining solution quality due to preselected index candidates.

\ SWIRL [6] represents a state-of-the-art index selection method that surpasses both traditional and RL-based approaches by incorporating a detailed workload model and action masking rules, effectively supporting multi-attribute indexes and excelling in generalizing to new query types. However, SWIRL’s sophistication comes with challenges, such as high training costs and complexity. Its detailed approach requires significant computational resources and expertise, and its dependence on manually defined pruning rules can limit training efficiency and adaptability in highly variable environments, highlighting the need for further research to improve its practicality and training process.

\ In conclusion, while traditional and early RL-based methods have laid the groundwork for automated index selection, they often fail to address the complexity of workloads and database environments fully. Our work seeks to fill this gap, offering a comprehensive solution that balances the need for efficiency, adaptability, and high-quality index configurations.

\

:::info Authors:

(1) Taiyi Wang, University of Cambridge, Cambridge, United Kingdom ([email protected]);

(2) Eiko Yoneki, University of Cambridge, Cambridge, United Kingdom ([email protected]).

:::

:::info This paper is available on arxiv under CC BY-NC-SA 4.0 Deed (Attribution-Noncommercial-Sharelike 4.0 International) license.

:::

\