Adaptive distributed traffic simulation with virtual partitioning in SUMO

Microscopic traffic simulations are a key enabler of Intelligent Transportation Systems (ITS) and urban Digital Twins (DTs), supporting both strategic planning and near–real-time traffic monitoring and optimization. Achieving high fidelity and low latency at scale, however, remains challenging due to the computational demands of simulating individual vehicle dynamics and the limited scalability of traditional monolithic execution models. Distributed and parallel simulation approaches offer a promising solution but are often affected by workload imbalances arising from the spatiotemporal evolution of traffic, which can significantly reduce resource efficiency. This paper proposes a distributed traffic simulation framework based on virtual network partitioning, in which parallel simulation instances maintain a full representation of the road network while actively simulating only assigned parts. This full-replica design trades higher per-node memory for low-cost runtime reconfiguration, enabling an adaptive load-balancing mechanism that monitors computational load and triggers repartitioning when when imbalances exceed a configurable threshold. While remaining applicable and portable across different simulators, the architecture is implemented and evaluated as an open-source extension of the SUMO microscopic traffic simulator, which does not natively support parallel execution. Experimental results on real-world urban networks demonstrate the performance of our system in small-to-medium deployments with speedups exceeding 2 × and traffic metrics preserved up to a minor bias. Furthermore, our study shows that adaptive load-balancing recovers the parallel advantage that would otherwise be lost under uneven traffic distributions, provided its configuration parameters are properly tuned, indicating that careful parameter selection is essential to sustain performance across heterogeneous ITS scenarios.