GraphRay: Distributed pathfinder network scaling

Pathfinder network scaling is a graph sparsification technique that has been popularly used due to its efficacy of extracting the “important” structure of a graph. However, existing algorithms to compute the pathfinder network (PFNET) of a graph have prohibitively expensive time complexity for large graphs: O(n
³
) for the general case and O(n
²
log n) for a specific parameter setting, PFNET(r = ∞, q = n - 1), which is considered in many applications. In this paper, we introduce the first distributed technique to compute the pathfinder network with the specific parameters (r = ∞ and q = n - 1) of a large graph with millions of edges. The results of our experiments show our technique is scalable; it efficiently utilizes a parallel distributed computing environment, reducing the running times as more processing units are added.