Automated Microservice Decomposition Method as Multi-Objective Optimization

IT architects expend a great deal of effort to manually decompose a current system into microservices (MSD) to improve the system maintainability. There are several methods that decompose the current system by automatically generating MSD candidates (MSDCs). These MSDCs are evaluated by defined evaluation functions, and the weighted sum of the evaluation is used to determine the best MSDC. However, it is difficult for IT architects to define the weighting on the spot. In this research, we propose an automated MSD method to generate a number of MSDCs by executing MSD as a multi-objective optimization problem. In the proposed method, a sufficient number of MSDCs is generated without weighting by using predefined MSD policies and fix operations. The generated MSDCs are mapped to a vector space that consists of basis vectors formulated in accordance with the evaluation functions defined by the MSD policies. Pareto-optimal solutions are then extracted from the mapped MSDCs by using a non-dominated sorting algorithm. We also extract MSDCs that are closest to the reference lines, which are defined to divide the vector space evenly, as distinctive MSDCs. We applied our method to two cases and found that it can automatically generate a sufficient number of distinctive MSDCs, thus enabling IT architects to efficiently find the best MSDC and rapidly execute MSD.