A Complexity Metric for Microservices Architecture Migration

Monolith applications tend to be difficult to deploy, upgrade, maintain, and understand. Microservices, on the other hand, have the advantages of being independently developed, tested, deployed, scaled and, more importantly, easier to change and maintain. This paper addresses the problem of migrating a monolith to a microservices architecture. Therefore, we address two research questions: (1) Can we define the cost of decomposition in terms of the effort to redesign a functionality, which is implemented in the monolith as an ACID transaction, into several distributed transactions? (2) Considering several similarity measures between domain entities, which provide a better decomposition when they are compared using the proposed complexity metric? To answer the first research question, we propose a complexity metric, for each functionality of the monolith application, that measures the impact of relaxing the functionality consistency on the architecture redesign and implementation. Regarding the second research question, we experiment with four similarity measures, each based on a different type of information collected from monolith functionality implementation. We evaluated our approach with three monolith systems and compared our complexity metric against industry metrics of cohesion and coupling. We also evaluated the different similarity measures in terms of the complexity of the decomposition they produce. We were able to correctly correlate the complexity metric with other metrics of cohesion and coupling defined in other research and we conclude that no single combination of similarity measures outperforms the other, which is confirmed by the existing research. Additionally, we conclude that the approach can help on an incremental migration to microservices, which, actually, is the strategy proposed by the industry experts.