A simple queue model for traffic flow was investigated. The main difference to queueing theory and to some dynamic extensions of static assignment is a hard storage constraint on the link; this generates spill-back. In contrast, the queue model is still simpler than the flow models of DYNASMART, DynaMIT, and the cell transmission model: The queue model does not divide links into segments. The disadvantage of this is that jam wave speeds are no longer realistically modeled; the advantage is higher computational performance.
Hard storage constraints have the consequence that link outflows are constrained both by the flow capacity of the link itself and by space limitations on the receiving link. In contrast to earlier versions of the queue model, we implemented ``fair'' intersections, where, if space on the outgoing links is limited, that space is allocated proportional to the incoming links' capacity.
The intention of this work is threefold: (1) Investigate minimal extensions of static assignment. (2) Investigate parallel computing limitations of transportation simulations in general. (3) Use the queue model as a simpler and faster alternative to TRANSIMS-like micro-simulations in activity-based transportation modelling.
With respect to computing, it was demonstrated that affordable Beowulf clusters (clusters of Pentium computers with Linux operating system) can be used successfully for large scale problems. The latency of Ethernet communication sets a hard limit on computing speed to about 150 simulation time steps per second, no matter what the problem size. It was shows that the use of Myrinet communications technology overcomes that problem, and at no higher cost, since with Myrinet one can reach better computing speeds with a smaller numbers of computers than with Ethernet. A speed-up of 180 was reached on 64 CPUs, meaning in practical terms that a month of computing can be reduced to about a day.
The model was applied successfully to a dynamic traffic assignment simulation of car traffic of the whole country of Switzerland for the morning peak. The results of this are reported elsewhere (37,38), including comparison to a VISUM assignment result and to field volume data.