A Graph-Based Dependency Analysis Method for Identifying Critical Components in Distributed Publish-Subscribe Systems

Modern distributed publish-subscribe (pub-sub) architectures enable loosely coupled, scalable communication among components, facilitating seamless integration and interoperability across diverse systems from IoT platforms to microservices. However, their complexity poses significant challenges in understanding system-wide dependencies and predicting the impacts of cascading failures. This paper introduces a comprehensive multi-layer graph-based analysis framework that captures both application-level and infrastructure-level dependencies in pub-sub systems, providing actionable insights for reliability engineering. Our methodology models distributed systems as directed multi-layer graphs, where applications, brokers, topics, and physical nodes are represented as vertices. At the same time, their relationships, including message publication, subscription, routing, and hosting, are depicted as typed directed edges. Beyond traditional graph analysis, we introduce a criticality score for each component to quantify its importance. We demonstrate how graph algorithms and visualizations can reveal critical components, identify single points of failure, and assess the impact of failures on service availability and performance. The case studies' results highlight the graph model's effectiveness in identifying critical components, enhancing the system's resilience, and guiding operators toward more robust configurations. The framework scales efficiently to systems with over 1000 components, while supporting efficient offline analysis and potential incremental updates. As a result, our graph-based dependency analysis approach offers immediate practical value for designing, maintaining, and evolving scalable, fault-tolerant pub-sub infrastructures in production environments.