Design and implementation of a parallel queue-based traffic flow simulation

Today, agent based micro-simulations are widely used in the field of transport planning and traffic management. One important requirement is the ability to simulate large scale scenarios in reasonable time. An obvious approach to reduce the computation time of such scenarios is to use multiple CPU cores. This paper presents the implementation of a parallel queue simulation for MATSim written in Java. Existing parallel traffic micro-simulations are reviewed concerning their parallelization approaches as well as the reached performance gains. Various concepts how to model12the progress of time and how to distribute computational workload among multiple CPU cores are discussed. Based on an analysis of the MATSim framework regarding its structure, performance and extensibility the concepts for the parallel queue simulation are selected and implemented. Performance tests with different sized scenarios are conducted. An analysis of the results shows that especially for large scale scenarios a significant performance gain is reachable