Microscopic simulation of traffic in networks: Supercomputer experience

The use of microscopic traffic-simulation codes in practice and in research has generally been limited by available computational resources to very small portions of an urban area’s network. Supercomputers offer an order-of-magnitude improvement over conventional mainframes, and thus provide the possibility of using microscopic simulation for much more elaborate and realistic simulations of traffic in large networks. This paper presents the results of computational experiments conducted with the NETSIM microscopic simulation package on a CRAY X-MP/24 supercomputer. The application to a large network of the central core of Austin, Texas is demonstrated. Comparisons of the computational performance of NETSIM on the supercomputer with that on a mainframe, for a set of test networks, are presented, confirming the magnitude of the improvements generally claimed. In addition, a relation is developed between the time requirements on the supercomputer and the characteristics of the network. Modifications to better vectorize the NETSIM code are identified. The implications for further research in traffic theory and its engineering applications are also discussed.