Efficient Parallel Shortest Path Algorithms

Abstract

Finding the shortest path between nodes in a graph has wide applications in many important areas such as transportation and computer networks. However, the current reference algorithms for this task, Dijkstra’s for single threaded environments and $\triangle$-stepping for multi-threaded ones, leave performance and efficiency on the table by not taking advantage of additional information available about the graph. In this paper we present and experimentally evaluate novel algorithms $SP_{1},SP_{2}$ and ParSP2 that leverage these constraints to solve the problem faster and more efficiently in key metrics. In single threaded execution, we show how SP1 and SP2 out-perform Dijsktra’s algorithm by up to 46%. In multi-threaded execution we show how our algorithms compare favorably to $\triangle$-stepping algorithm in the ability to establish the shortest path between the source and the median node.