Empirical Comparison of Runtime Improvement Approaches: Genetic Improvement, Parameter Tuning, and Their Combination

Software can be optimised in various ways, for instance, by modifying its source code or adjusting its compiler and runtime parameters. To automate these tasks, algorithm configuration and genetic improvement have been proposed the former modifies parameters and the latter source code. Many tools have been introduced to automate such changes. However, these tools typically only work at a single code level, optimising either parameter values or source code, but not both. In 2022, Blot and Petke [1] introduced MAGPIE, a framework capable of simultaneously searching for improvements at multiple granularity levels. Our literature review revealed that the best search strategies in genetic improvement and algorithm configuration that may generalise to both domains are based on local search and genetic algorithms, respectively. We compared these two approaches for improving the execution time of the MiniSAT solver, and also explored their performance on the joint search space of parameter and source code edits. Our results show that genetic improvement with first improvement local search led to the best results, improving MiniSAT's execution time by 18.05%.