Toward Efficient Asynchronous Single-Source Shortest Path

Marco D'Antonio, Son Thai Mai, Hans Vandierendonck

Abstract: Single-Source Shortest Path (SSSP) is a fundamental graph problem that arises in various applications and complex problems. State-of-the-art solutions to the parallel SSSP problem create parallelism through priority coarsening, which results in redundant work and reduces the efficiency of the solution. This paper introduces Wasp, a novel solution for SSSP that addresses the parallelism-redundant work problem using an asynchronous work-stealing scheduler. The experimental evaluation of Wasp on SSSP shows that it provides competitive or better performance than GAP, GBBS, and the MultiQueue on 13 diverse graphs, including road networks, scale-free and random graphs, with speedups up to 2.94× higher than the state of the art.

External IDs:dblp:conf/ipps/DAntonioMV25