AI 中文总结
本文提出一种微服务识别工具的统一特征模型，经评估其可捕获现有方法的主要变体点，现有工具虽未覆盖模型大部分内容，但共同覆盖几乎所有变体点，为微服务识别工具的分析、比较与设计提供了统一基础。
AI 中文摘要
已有多种针对单体系统内微服务自动识别的方法被提出，这些方法在数据收集与分析技术、所应用的分解算法，以及用于可视化和优化候选微服务的机制上存在差异。尽管存在这种多样性，但由于缺乏通用的概念框架，不同方法间的系统实验与比较仍受限。此外，当前研究表明不存在单一最优方法；相反，需整合多种方法以充分探索任何设计方案固有的权衡。为解决这一缺口，本文在对现有技术的广泛分析基础上，提出一种适用于变体丰富的微服务识别工具的特征模型。我们通过对代表性文献的系统映射，以及分析该模型在现有微服务识别工具架构中的实例化来评估此特征模型。研究结果有两点：其一，所提出的特征模型成功捕获了现有方法的主要变体点，为分析、比较和设计微服务识别工具提供了统一基础；其二，虽然没有单个工具覆盖该模型的大部分内容，但被分析的工具共同覆盖了几乎所有变体点。这种互补性是核心结果：设计空间已被占据，但它分散在各个独立工具中，没有任何工具能比较或整合其他工具所实现的替代方案。
英文摘要
Several approaches have been proposed for the automatic identification of microservices within monolithic systems. These methodologies differ in their data collection and analysis techniques, the decomposition algorithms applied, and the mechanisms used to visualize and refine candidate microservices. Despite this diversity, systematic experimentation and comparison across approaches remain limited by the absence of a common conceptual framework. Furthermore, current research indicates that no single optimal method exists; rather, integrating multiple approaches is necessary to fully explore the trade-offs inherent in any design solution. To address this gap, this paper proposes a feature model for variant-rich microservice identification tools, grounded in an extensive analysis of the state of the art. We evaluate this feature model through a systematic mapping of representative literature and by analyzing its instantiation within the architecture of an existing microservice identification tool. Our findings are two-fold. First, the proposed feature model successfully captures the primary variation points of existing methodologies, providing a unifying foundation for analyzing, comparing, and designing microservice identification tools. Second, while no individual tool covers more than a fraction of the model, the analyzed tools jointly span almost all of its variation points. This complementarity is the central result: the design space is already populated, but it is fragmented across standalone tools, none of which can compare or integrate the alternatives that the others implement.
Comments26 pages, 17 figures and 13 tables
