Designing a Time-Driven Simulation Framework for Large-Scale Traffic Networks

In the field of traffic simulation, the shift towards large-scale, time-driven models marks a significant departure from traditional event-driven mechanisms, necessitating robust traffic dynamics that accurately capture the complexities of urban mobility. This study introduces a time-driven simulation framework designed to rigorously evaluate the performance of multi-GPU parallel computing systems across extensive traffic networks. Incorporating a traffic network graph with 224,223 nodes and 549,008 edges, alongside a comprehensive traffic demand of 24 million trips, this benchmark establishes a new standard for assessing simulation efficiency, scalability, and reproducibility. Through the Large Scale Multi-GPU Parallel Computing based Regional Scale Traffic Simulation Framework (LPSim), we achieve notable advancements in simulating dynamic traffic networks with unmatched speed and accuracy, significantly surpassing traditional CPU-based methods. We tested our simulator with 9.01 million demand trips with the network described above on dual NVIDIA A100-PCIE-40GB GPUs, which finished the simulation with 0.0398483333 hours simulation time, which is around 113 times faster than the same simulation scenario running on an Intel(R) Xeon(R) Gold 6326 CPU @ 2.90GHz, which costs 4.49192555556 hours to finish. LPSim’s versatility across multi-modal contexts and configurations provides a robust platform for future innovations in traffic simulation frameworks, bridging the theoretical models and practical urban planning applications. The code is available at: https://github.com/Xuan-1998/LPSim