Dependencies-based microservices decomposition method

Abstract

A microservices identification method was proposed by this research paper. The proposed method consists of two parts; the first part is representing the source code of the monolithic application as a class dependency graph. This graph represents the structure of the monolithic application and the relationships between the classes of the application. The second part of the method is a graph clustering algorithm to identify the microservices through analyzing the dependencies between the classes of the monolithic application and cluster classes with solid relationships to generate microservice candidates. The method was tested with 8 different applications and 11 clustering algorithms were examined to find the most accurate and efficient algorithm. The proposed method produced promising results when compared to other methods in the literature with 0.8 averaged F-Measure ‘F1’ score and 0.44 averaged NGM score. The F1 score shows that the proposed method has good accuracy in detecting microservices candidates. Newman Girvan Modularity metric ‘NGM’ score shows that the generated microservices candidates are properly structured and that there are well-defined relationships among the clustered classes of the generated microservices.