Dynamic and Static Feature-Aware Microservices Decomposition via Graph Neural Networks

Dynamic and Static Feature-Aware Microservices Decomposition via Graph Neural Networks

연구 분야: Software Development

학회: International Conference on Knowledge Science, Engineering and Management

Microservices architecture has gained significant success in software development due to its flexibility and scalability. Decomposing a monolithic system into microservices can increase code reusability and reduce reconstruction costs. However, existing microservices decomposition approaches only utilize dynamic or static feature to represent the monolithic system, leading to low coverage of classes and inadequate information. To address these issues, we propose a novel Microservices Decomposition approach with the Dynamic and Static feature-aware called MDDS, which combines dynamic and static feature to construct a graph that represents the monolithic system. We also propose an approach based on variational graph auto-encoders to learn the nodes’ representations in the graph, providing similar encodings to functionally similar classes to enhance microservices clustering. Extensive experiments on four traditional monolithic systems demonstrate that MDDS outperforms other baselines in several significant metrics.