基于运行时特征的单体系统微服务拆分方法研究  

Research on Microservice Splitting Method of Monolithic System Based on Runtime Characteristics

作　　者：沈瑞娜 陈璟 王春生[2] 赵迎泽 张鹏[2] SHEN Ruina;CHEN Jing;WANG Chunsheng

机构地区：[1]航空工业第一飞机设计研究院,西安710089 [2]西北工业大学软件学院,西安710129

出　　处：《科技创新与应用》2025年第12期15-19,共5页Technology Innovation and Application

基　　金：国防科技173计划(2022-JCJQ-JJ-0581);陕西省重点研发计划(S2023-YF-YBGY-0279)。

摘　　要：受益于微服务架构所带来的优势,业界众多开发团队已经逐渐将单体架构软件向微服务架构迁移。目前,关于微服务拆分方法的研究存在着依赖人工经验、缺乏考虑软件运行时动态特征等问题。为此,设计基于运行时特征的微服务拆分方法。算法采用两阶段拆分的策略,以动态数据为主,静态数据为辅,数据层面边界清晰,并在算法层面对数据进行融合。在静态调用矩阵和动态调用关系矩阵上分别做层次聚类,将所得2个矩阵融合,使用谱聚类算法得到功能原子集,进而使用NSGA-II算法对功能原子集进行划分,得到最终的微服务拆分方案。实验结果表明,提出的方法能有效提高微服务拆分质量,得到的拆分结果更为合理可靠。Benefiting from the advantages offered by microservice architecture,many development teams in the industry have gradually migrated from monolithic to microservice-based architectures.However,current research on microservice decomposition methods faces challenges such as reliance on manual expertise and the lack of consideration for dynamic runtime characteristics of the software.To address these issues,we propose a microservice decomposition method based on runtime features.The algorithm adopts a two-phase decomposition strategy,primarily using dynamic data with static data as a supplement,ensuring clear data-level boundaries.The data is integrated at the algorithmic level.Hierarchical clustering is performed on both the static call matrix and the dynamic call relationship matrix,and the two matrices are then fused.A spectral clustering algorithm is used to obtain a set of functional atoms,which are subsequently divided using the NSGA-II algorithm to produce the final microservice decomposition plan.The experimental results show that the proposed method can effectively improve the quality of microservice splitting,and the splitting results obtained are more reasonable and reliable.

关 键 词：微服务拆分 软件架构迁移 微服务架构 软件系统 内部特征 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]