A new framework based on features modeling and ensemble learning to predict query performance.

Area of Science:

• Computer Science
• Database Systems
• Machine Learning

Background:

• Query optimizers traditionally require significant overhead for performance statistics.
• Existing query performance prediction models often rely on database management system (DBMS) statistics and operator cost estimation.
• Accurate modeling of query features is crucial for robust performance prediction.

Purpose of the Study:

• To propose a novel framework for predicting query performance using query feature modeling and ensemble learning.
• To develop a query performance predictor simulator to identify influential query features.
• To optimize query features that impact query performance.

Main Methods:

• Developed a framework incorporating query feature modeling across five dimensions: syntax, hardware, software, data architecture, and historical performance logs.
• Utilized ensemble learning for the performance prediction model, adept at handling missing values and overfitting via regularization.
• Created training datasets from real-world performance data logs for model training and validation.

Main Results:

• The proposed framework effectively models query features influencing performance.
• Ensemble learning demonstrated robustness in handling data complexities like missing values and overfitting.
• Experimental results showed the proposed prediction model's effectiveness compared to existing approaches.

Conclusions:

• The novel framework provides an effective approach to query performance prediction.
• Query feature modeling combined with ensemble learning offers a powerful solution for optimizing database performance.
• Empirical evidence supports the superiority of the proposed method over related work.