Analyzing per-component energy consumption of evolutionary algorithms implemented in low-level languages: Comparison of C++ and zig in key evolutionary algorithm functions

Trying to design software systems in such a way that they consume a minimal amount of energy is simply good engineering practice. Scientific software is no exception; in this case, we will put our focus on evolutionary algorithms and the energy consumption patterns in memory and CPU for two different, low-level languages: The mainstream C++ and the emerging zig. By setting up a methodology that gives us a precise measure of the energy spent by key evolutionary algorithm functions, we can give the scientific software engineer some actionable insights on how to write energy-conscious, and thus energy-thrifty, evolutionary algorithms. Our experiments show that, even with very low energy consumption in both cases, C++ can achieve a significant reduction in energy consumption for some integer-based fitness functions, as well as very good performance on classical genetic operators. Besides, the experimental results have a low variability, as compared to zig, making it in the short and medium run the best of the two languages for evolutionary algorithms.