Vehicular Traffic Networks Simulation: Towards the Implementation of a High-Performance Computing Simulator

This work presents a comparison of different software and hardware implementations of a traffic simulator using a fluid-based mesoscopic model. The model represents Vehicular Traffic Networks (VTNs) based on the flow resulting by the interconnection of dynamical agents associated to street segments and intersections. Starting from a single cross-intersection example, a traditional numerical simulation was developed in MATLAB as a first step. Afterwards looking for a leverage of the current computer capabilities, new approaches in software and hardware were explored. Using the Python language and Django as a development framework, the next version of the simulator was built to exploit the high-performance computing capabilities for research with a web interface. Next, using a modular paradigm, a second version of a simulator was made based on object-oriented programming (OOP) having “high-cohesion”, “low-coupling”, and extensibility features. A third numerical version with some libraries like NumPy was created and tested, looking for better performance. All these versions served as a point of departure to explore the basis of a fourth and fifth version: a high-performance simulator with a parallel mindset, using Graphic Processing Units (GPUs). A comparison is presented with the same single cross example testing the time to generate data in each version.