High Performance Computing applied to Dynamic Traffic Assignment

Abstract

This paper aims at applying High Performance Computing to Dynamic Traffic Assignment. The latter are a class of algorithms designed for accurate traffic simulations on networks. The motivation to bring together these two distinct fields of study is primarily to enable traffic engineers to tackle real-size traffic problems on large networks, using the powerful approach of Dynamic Traffic Assignments. Various insights and guidelines are presented, as well as their impacts on computation time and memory footprint. The main contribution of this paper is the method of domain separation, which is applied to cut a road network into regions. It enables parallel computing in a cluster environment, whilst ensuring a consistent synchronization across the global network. Additional speed ups are achieved by balancing the amount of operations within a region and the quality of the overall result. It is hoped that this study can bring interest for researchers of both domains, to further improve the efficiency of traffic algorithms for realistic simulations.