Divide, Conquer, and Match: A Distributed and Asynchronous Approach for Subgraph Isomorphism

Subgraph isomorphism, the task of identifying whether a smaller pattern graph exists within a larger data graph, is a cornerstone problem in graph analytics with applications spanning bioinformatics, social network analysis, and cheminformatics. Despite its utility, the NP-complete nature of subgraph isomorphism poses significant computational challenges, especially with large-scale graphs. This paper presents a distributed and asynchronous approach for subgraph isomorphism using an actor-based approach. Our method employs recursive decomposition of pattern graphs into enriched subpatterns, followed by fine-grained distributed parallel pattern matching. By overlapping computation and communication through lightweight asynchronous active messages, our approach achieves notable speedups compared to related task-based and migrating thread-based implementations. Evaluations on the IARPA AGILE program’s dataset and provided reference implementations demonstrate improved performance and scalability, achieving up to 11.6× speedup.