From Everywhere to Everywhere (FETE): Adaptation of a Pedestrian Movement Network Model to a Hybrid Parallel Environment

Shortest-path algorithms are hard to parallelize because they require a large number of global operations to estimate the costs of alternative routes. However, some geographic problems, such as locating archaeological sites and tracking the spread of infectious diseases, demand the ability to find a large number of the shortest paths on very large graphs or grids. Here, we present an approach based on the out-of-RAM Dijkstra shortest-path algorithm that can be employed in hybrid massively parallel or cloud environments. In this approach, we partition the graph, precompute all paths inside each partition, and then assemble the routes from precomputed paths. We demonstrate the utility of this approach by estimating travel frequency in pedestrian networks.