Service Candidate Identification from Monolithic Systems Based on Execution Traces

• Published in: IEEE transactions on software engineering Vol. 47; no. 5; pp. 987 - 1007
• Main Authors: Jin, Wuxia, Liu, Ting, Cai, Yuanfang, Kazman, Rick, Mo, Ran, Zheng, Qinghua
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.05.2021; IEEE Computer Society
• Subjects: Algorithms; Atomic measurements; Candidates; Complexity; Computer architecture; Evaluation; evolvability; execution trace; Frequency measurement; functionality; History; Maintainability; Microservice; Modularity; monolith decomposition; service candidate; Software; Testing
• ISSN: 0098-5589; 1939-3520
• DOI: 10.1109/TSE.2019.2910531

Monolithic systems increasingly suffer from maintainability and scalability issues as they grow in functionality, size, and complexity. It is widely believed that (micro)service-based architectures can alleviate these problems as each service is supposed to have the following characteristics: clearly defined functionality , sufficient modularity , and the ability to evolve independently . Industrial practices show that service extraction from a legacy monolithic system is labor-intensive and complex. Existing work on service candidate identification aims to group entities of a monolithic system into potential service candidates, but this process has two major challenges: first, it is difficult to extract service candidates with consistent quality; second, it is hard to evaluate the identified service candidates regarding the above three characteristics. To address these challenges, this paper proposes the Functionality-oriented Service Candidate Identification ( FoSCI ) framework to identify service candidates from a monolithic system. Our approach is to record the monolith's execution traces, and extract services candidates using a search-based functional atom grouping algorithm. We also contribute a comprehensive service candidate evaluation suite that uses interface information, structural/conceptual dependency, and commit history. This evaluation system consists of 8 metrics, measuring functionality, modularity, and evolvability respectively of identified service candidates. We compare FoSCI with three existing methods, using 6 widely-used open-source projects as our evaluation subjects. Our results show that FoSCI outperforms existing methods in most measures.