Reinforcement Learning Index Selection Under Storage Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing index selection methods for database systems struggle to balance storage consumption and performance gain efficiently, particularly for complex workloads, and fail to provide high-quality solutions quickly, especially in cloud environments with dynamically changing workloads.
Innovation Solution
A reinforcement learning-based index selection method using Proximal Policy Optimization (PPO) to iteratively determine an optimal set of indexes, incorporating workload representation and dynamically restricted action spaces, allowing for multi-attribute indexes and storage budget constraints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional iterative index selection algorithms are used, then high-quality index selection can be achieved, but the runtime becomes very long
Solution Approach 1:
The patent replaces traditional mechanical iterative algorithms with a reinforcement learning-based system. The RL agent learns to select indexes by interacting with the database workload, using neural networks to evaluate index candidates and prioritize actions. This substitution of the decision-making mechanism enables faster index selection while maintaining high quality through learned policies optimized for both performance and runtime.
Solution Approach 2:
The patent changes the fundamental parameters of the index selection process by introducing reinforcement learning components. Instead of traditional iterative evaluation, the system uses RL agents that learn from workload patterns, employing policies and value functions to make decisions. This parameter change in the algorithmic approach enables simultaneous optimization of solution quality and computation time.
2Extent of automation
If cloud database systems with dynamically changing workloads are deployed, then total cost of ownership is reduced, but the complexity of maintaining optimized configurations increases
Solution Approach 1:
The patent implements self-service through autonomous RL agents that automatically adapt index selections to changing workloads. The agents continuously learn from workload patterns and adjust index configurations without human intervention. This automation handles the complexity of maintaining optimized configurations in cloud environments, allowing the system to adapt dynamically while reducing operational burden.
Solution Approach 2:
The system employs feedback mechanisms where RL agents observe workload execution patterns and use this information to refine their index selection policies. The agents learn from actual query performance and adjust their decisions accordingly, creating a closed-loop system that automatically optimizes configurations in response to changing cloud workloads.
3Quantity of substance
If reductive index selection approaches are used, then storage consumption is reduced, but the runtime increases significantly due to many iterations
Solution Approach 1:
The patent applies preliminary action by having the RL agent learn and prioritize index candidates before actual selection occurs. The agent evaluates and ranks potential indexes based on predicted workload patterns, enabling faster selection without exhaustive iteration. This pre-computation of index priorities significantly reduces runtime while maintaining efficient storage consumption.
Data Source
AI summary
Embodiments relate to a method for training an index selection agent for iteratively determining an index set I to be used when running queries against a database, the method being implemented in a computer system. Preferably, the index selection agent is based on Reinforcement Learning.