Increasing Parallelism in Forward-backward Distributed Algorithm for Finding Strongly Connected Components of Directed Graphs

Abstract

The paper proposes a modification of the existing distributed forward-backward algorithm for finding strongly connected components in directed graphs. The modification aims at improving the parallelism of the algorithm by increasing the branching factor while dividing the workload. Instead of randomly picking the pivot vertex, a heuristic technique is used which allows more sub-tasks to be generated, on average, for the subsequent step of the algorithm. The work describes suitable algorithm modifications and presents empirical results, proving suitability of the approach in question.