From legacy to microservices: A type-based approach for microservices identification using machine learning and semantic analysis

The microservices architecture (MSA) style has been gaining interest in recent years because of its high scalability, ability to be deployed in the cloud, and suitability for DevOps practices. While new applications can adopt MSA from their inception, many legacy monolithic systems must be migrated to an MSA to benefit from the advantages of this architectural style. To support the migration process, we propose MicroMiner, a microservices identification approach that is based on static-relationship analyses between code elements as well as semantic analyses of the source code. Our approach relies on machine learning (ML) techniques and uses service types to guide the identification of microservices from legacy monolithic systems. We evaluate the efficiency of our approach on four systems and compare our results to ground-truths and to those of two state-of-the-art approaches. We perform a qualitative evaluation of the resulted microservices by analyzing the business capabilities of the identified microservices. Also a quantitative analysis using the state-of-the-art metrics on independence of functionality and modularity of services was conducted. Our results show the effectiveness of our approach to automate one of the most time-consuming steps in the migration of legacy systems to microservices. The proposed approach identifies architecturally significant microservices with a 68.15% precision and 77% recall.