Monolith to Microservices: VAE-Based GNN Approach with Duplication Consideration

Korn Sooksatra, Rokin Maharjan, Tomás Cerný

Abstract: With the rise of cloud computing, many applications have been implemented into microservices to fully utilize cloud computing for scalability and maintainability purposes. However, there are some traditional monolith applications that developers would like to partition into microservices. Unfortunately, it is difficult to find a solution when considering multiple factors (i.e., the strong dependency in each cluster and how often different microservices communicate with each other). Further, because we allow duplications of classes in multiple microservices to reduce the communications between them, the number of duplicated classes is also another important factor for maintainability. Therefore, we need to use machine learning algorithms to approximate a good solution due to the infeasibility of finding the optimal solution. We apply the variational autoencoder to extract features of classes and use the fuzzy c means to group the classes into microservices according to their extracted features. As a result, our approach outperforms the other baselines in some significant metrics. Also, when we allow duplication, we find that it is helpful in terms of reducing the overhead of communications between microservices.