Time-dependent shortest path problem using MapReduce

Searching for the fastest path in a road network is influenced by many temporally changing circumstances such as a traffic density, weather conditions or road constructions. Even though many of these factors can be predicated in advance, most of the currently used solutions searching for a shortest path does not consider temporal traffic condition changes. In this paper we focus on problem of searching a shortest path in a large graph where the cost of the edge is time dependent. In particular we focus on the selection of the best time of departure and the shortest path calculation to achieve the fastest travelling time from one node to all other nodes of the graph. To be able to find such information in a large graph exceeding a commodity hardware parameters, we will focus on distributed computing environment based on MapReduce model. We test and evaluate a MapReduced algorithm derived from Label Correcting algorithm with respect to data size, number of cluster nodes, algorithm complexity as well as the result correctness. Moreover we applied the method on searching a time dependent path in a real word road network by using OpenStreetMap data and randomly generated time dependent edge cost functions. Finally a description of an implementation based on Apache Spark and GraphX framework is given and the benchmarking results measured on a production cluster are mentioned.