Parallel Shortest Path Algorithms : Identifying the Factors thatA ect

Shortest path computation is required by a large number of applications such as VLSI, transportation and communication networks. These applications, which often use parallel processing, require an eecient parallel shortest path algorithm. The experimental work related to parallel shortest path algorithms has focused on the development of eecient, parallel label-correcting or label-setting algorithms; however, no in depth analysis of the diierence between the parallel algorithms has been made. In this paper we conduct a thorough experimental analysis of parallel shortest path algorithms for sparse networks, identifying the factors that aaect the performance of these parallel algorithms for distributed memory machines: number of total updates, work per update and number of iterations. The results of our analysis indicates that our variant of the label-setting algorithm has better parallel performance than label-correcting, especially for large networks. This occurs because of timing issues; the label-setting algorithms receive more messages on average per iteration than label-correcting due to smaller variances in local compute times.