An Efficient Parallel Ordered Depth-First Search Strategy for Directed Acyclic Graphs

Abstract

With the advent of the big data era, accelerating the parallelization of Depth-First Search (DFS) has become pivotal for addressing the challenges posed by large-scale datasets and complex problems in contemporary applications. To improve the parallel processing performance of DFS on Directed Acyclic Graphs (DAGs) while maintaining the orderliness of traversal outcomes, this paper introduces an efficient Parallel Ordered Depth-First Search (PODFS). By leveraging the novel concepts of Clue Path, ParallelList, and the node attributes Level and Dis, PODFS achieves precise subgraph partitioning while preserving the ordered nature of parallel search results. After performing a one-time preprocessing on a specific graph, the proposed algorithm enables more efficient global traversals, achieving a speedup ranging from 6× to 12× on various real-world graph datasets while maintaining the orderedness of traversal results. These performance improvements are crucial for applications that require frequent, in-depth graph searches with a strict need to preserve traversal order.