This work aims to present a new event-driven queue-based approach that makes the traffic flow microsimulation of large-scale problems feasible on affordable, desktop computer hardware within a reasonable time. RELATED WORK This section presents a selection of previous work related to the microsimulation. There are many different traffic flow simulation approaches. The physically based microsimulations [e.g., AIMSUN (1, 2), MITSim (3, 4), and VisSim (5)] generally try to capture as many traffic flow phenomena as possible. They simulate car-following and lane-changing behavior and use a continuous representation of space and constant, very small time steps to simulate cars on the roads. A different microscopic, but less physical, simulation approach is represented by cellular automata (6), used, for example, in TRANSIMS (7–9). Here, cars move through cells that they can occupy like par- ticles. Although there is a coarser level of detail in cellular automata, features such as densities and travel speeds still emerge from the cars’ simple direct interactions and are not computed at an aggregated level. This changes when one moves to mesoscopic modeling. In mesoscopic models [e.g., Metropolis (10, 11), DynaMIT (12, 13), Dynasmart (14, 15), DYNEMO (16), and ORIENT/RV (17 )], where vehicles are still represented microscopically, travel times and speeds are calculated by using aggregates. The highest level of abstraction can be found in macroscopic models that compute all traffic quantities on an aggregated level. One example of a traffic simulation model as a one-dimensional incompressible fluid is Netcell (18). The work presented here builds on the approach of MatSim-T (19, 20). This model uses simplified, queue-based dynamics to be computationally efficient but still estimates all quantities microscop- ically and thus should be located somewhere between mesoscopic approaches and cellular automata. The same holds for the approach presented by Mahut (21) where the spatial resolution on the links is minimized by calculating only the entry and exit times of vehicles. Time-step-based approaches have been more popular in the past than event-driven approaches. It is not clear why, but one reason might be the more straightforward implementation of time-step-based simulations. CLASSIFICATION OF TRAFFIC FLOW MICROSIMULATION METHODS Cellular Automata Cellular automata are used in traffic flow microsimulation to model accurately the behavior of cars traveling on a road network [see, e.g., Chowdhury et al. (22)]. The basic idea is to discretize space in Event-Driven Queue-Based Traffic Flow Microsimulation David Charypar, Kay W. Axhausen, and Kai Nagel 35 Simulating traffic flow is an important problem in transport planning. The most popular simulation approaches for large-scale scenarios today are aggregated models. Unfortunately, these models lack temporal and spatial resolution. But microsimulations are interesting for traffic flow simulation, as they can accurately simulate features requiring both high temporal and high spatial resolution, including traffic jams and peak periods. However, most microscopic approaches involve high computa- tional costs and expensive large computers to run them within a reason- able time. This work presents possibilities for reducing these costs by using a queue-based model and an event-driven approach jointly. The approach makes it possible to run large-scale scenarios with more than 1 million simulated person-days on networks with 10,000 links in less than 10 min on single CPU desktop computers present in most offices today. Traffic flow simulation is an important problem in transport planning, as it represents the final link between the intangible description of travel demand and the emergence of flow densities, volumes, and travel speeds. In today’s practice, traffic flow is most often simulated by using aggregated models that are easy to use and well established in the community. However, compared with aggregated models, traffic flow microsimulation has certain clear advantages. • Spatial and temporal resolution are high. Features like traffic jams and peak periods can be captured much more accurately. • Traffic is represented in a natural way; vehicles traveling, roads, and intersections are simulated directly. • Traffic flow microsimulations can be easily coupled with other microscopic approaches (i.e., agent-based demand modeling). • Inverse analyses can be carried out to determine where certain cars are coming from and why they can be found at a specific road network location. These properties come with a high price of computational burden, which often makes it necessary to use complex software architectures together with large parallel computers—an expensive process. The other option is to use microsimulations only for small applications and switch to the previously mentioned aggregated methods for large- scale problems. Unfortunately, these methods are so different that it is nearly impossible to transfer know-how from one field to another. D. Charypar and K. W. Axhausen, Institute for Transport Planning and Systems, Swiss Federal Institute of Technology, Zurich, Switzerland. K. Nagel, Transport Systems Planning and Transport Telematics, Technical University of Berlin, Germany. Corresponding author: D. Charypar, charypar@ivt.baug.ethz.ch. Transportation Research Record: Journal of the Transportation Research Board, No. 2003, Transportation Research Board of the National Academies, Washington, D.C., 2007, pp. 35– 40. DOI: 10.3141/2003-05