GPU-based Real-Time Execution of Vehicular Mobility Models in Large-Scale Road Network Scenarios Kalyan S. Perumalla, Brandon G. Aaby, Srikanth B. Yoginath, Sudip K. Seal Oak Ridge National Laboratory Oak Ridge, Tennessee, USA Abstract A methodology and its associated algorithms are presented for mapping a novel, field-based vehicular mobility model onto graphical processing unit computational platform for simulating mobility in large-scale road networks. Of particular focus is the achievement of real-time execution, on desktop platforms, of vehicular mobility on road networks comprised of millions of nodes and links, and multi- million counts of simultaneously active vehicles. The methodology is realized in a system called GARFIELD, whose implementation details and performance study are described. The runtime characteristics of a prototype implementation are presented that show real-time performance in simulations of networks at the scale of a few states of the US road networks. 1. Introduction 1.1. Motivation Simulations are routinely used in emergency planning and management in order to make decisions such as whether to order an evacuation or not [1, 2]. The quality of decisions can greatly depend on the quality of insights into simulation results. When larger geographical regions are considered in such decisions, simulations become highly computationally intensive. Improving the speed of large-scale simulation can help evaluate an increased number of alternatives, and improve confidence bounds, all within the short amount of decision time available. Large-scale scenarios of vehicular traffic simulation problems are characterized by long-range queuing effects, control mechanisms and other phenomena. While small-sized scenarios are relatively easy to analyze, larger scenarios need specialized treatment for efficient execution, especially for very large network sizes (millions of road intersections) and/or for heavy loads of vehicular traffic load (several million simultaneously active vehicle counts). An appealing computational platform in this context is a graphical processing unit (GPU). 1.2. Background Graphical processing units have been subjected to general-purpose use over the past decade. Literature on general-purpose computation over GPUs is extensive [3, 4]. However, newer methodologies and implementation approaches are still being discovered to exploit GPUs in different areas. Although GPU- based execution is not new, and the computational potential of GPUs has been known, no specific method has been proposed to map vehicular mobility models to GPUs. Traditional CPU-based (time-stepped or event- driven) models have remained elusive for straightforward application to the GPU domain. 1.3. Contributions To the best of our knowledge, ours is the first work to apply GPU-based model execution to transportation network simulations. Also, we are not aware of any other system or approach that has been shown to support queuing effects in either aggregate or semi- aggregate models of vehicular mobility at the level of millions of road network nodes and links. Ours is also the first to provide a novel field-based formulation of a vehicular traffic mobility model in a large road network that can be executed on a GPU. Modeling dynamic re-routing is another distinguishing aspect of our field-based model that has never been attempted before at large-scale in other models and simulators. 1.4. Related Work Commonly-used execution approaches span a continuous spectrum, between fully disaggregated, agent-based models, and fully abstracted, network flow analysis formulations. Examples of flow analysis- based methods include macro-simulators CORSIM [5, 6] and OREMS [1]. Examples of disaggregated approaches include micro-simulators such as TRANSIMS [7], VISSIM [8], and SCATTER[9], among many others. The literature on macro- simulators and micro-simulators in the mobility domain is extensive. The reader is referred to [1, 2, 5- 15] as starting points. However, field-based modeling