ABSTRACT
In this chapter we introduce the use of Graphical Processing Units (GPU) for multi-agents-based systems as an example of a not-so-regular application that could benefit from the GPU computing power. Multi-Agent Systems (MAS) are a simulation paradigm used to study the behavior of dynamic systems. Dynamic systems as physical systems are often modeled by mathematical representations and their dynamic behavior is simulated by differential equations. The simulation of the system thus often relies on the resolution of a linear system that can be efficiently computed on a graphical processing unit as shown in the preceding chapters. But when the behavior of the system elements is not uniformly driven by the same law, when these elements have their own behavior, the modeling process is too complex to rely on formal expressions. In this context MAS is a recognized approach to model and simulate systems where individuals have an autonomous behavior that cannot be simulated by the evolution of a set of variables driven by mathematical laws. MAS are often used to simulate natural or collective phenomena whose individuals are too numerous or various to provide a unified algorithm describing the system evolution. The agent-based approach is to divide these complex systems into individual self-contained entities with their smaller set of attributes and functions. But, as for mathematical simulations, when the size of the MAS increases, the need of computing power and memory also increases. For this reason, multi-agent systems should benefit from the use of distributed computing architectures. Clusters and grids are often identified as the main solution to increase simulation performance but GPUs are also a promising technology with an attractive performance/cost ratio.
