Best practices for energy-thrifty evolutionary algorithms in the low-level language zig

Abstract

The most fruitful way of making evolutionary algorithms spend the least amount of energy is to consider all possible program- ming techniques and platform choices that could, theoretically, affect performance, and carry out experiments using EA workloads in different platforms, eventually choosing those techniques that yield the minimum amount of energy expenses. These techniques include a choice of differ- ent data structures, as well as affecting compilation in such a way that energy footprint is reduced; they have to be replicated in different com- puting platforms because these expenditures may be affected by all the layers of the operating system and runtime framework used. In this paper we will experiment with different data structures and code refactoring techniques in the low-level language zig, trying to design rules of thumb that will help developers create green evolutionary algorithms. We will include two different hardware platforms, looking for the one that spends the least energy.