Core Concepts
Proposing a Meta Multi-Objectivization (MMO) model to optimize software configuration tuning by balancing target and auxiliary performance objectives.
Abstract
Software configuration tuning is crucial for optimizing performance objectives. The MMO model introduces two meta-objectives to prevent local optima traps while focusing on the primary objective. Normalization methods help balance the weight parameter sensitivity in the MMO model, improving efficiency and effectiveness.
Stats
Experiments on 22 cases from 11 real-world software systems/environments confirm that MMO outperforms single-objective counterparts on 82% of cases with up to 2.09× speedup.
For 68% of cases, the new normalization enables MMO to outperform prior work under pre-tuned best weights, saving resources.
The MMO model consolidates recent tuning tools on 68% of cases with up to 1.22× speedup.