Distributed intelligence in large scale traffic simulations on parallel computers

Kai Nagel¹
Dept. of Computer Science, ETH Zürich
CH-8092 Zürich, Switzerland

Abstract:

Transportation systems can be seen as displaying meta-intelligence, in the sense that intelligent actors (travelers) conspire to make the system function as a whole. In simulations one can model this by resolving each traveler individually, and giving each traveler rules according to which she/he generates goals and then attempts to achieve them. The system as a whole has no goal of its own except to ``function'', i.e. to be a metropolitan region where people's lives are good enough so they do not move away.

This paper approaches this question from an extremely pragmatic point - how can a large scale simulation of such a system be implemented on a parallel computer? In particular, we concentrate on Beowulf clusters, which are clusters of regular PCs connected by regular relatively slow local area network. Our working hypothesis is that the imbalance of the computational system -fast CPUs, slow communication- resembles the real system, and that a good simulation of the real system will take advantage of these parallels. The paper focuses both on what is actually implemented and working, and on future plans.