Multiprocessor Architectures Specialized for Multi-agent Simulation

Abstract

Two new multiprocessor architectures to accelerate the simulation of multi-agent worlds based on the massively parallel GCA (Global Cellular Automata) model are presented. The GCA model is suited to describe and simulate different multi-agent worlds. The designed and implemented architectures mainly consist of a set of processors (NIOS II) and a network. The multiprocessor systems allow the implementation in a flexible way through programming, thus simulating different behaviors on the same architecture. Two architectures with up to 16 processors were implemented on an FPGA. The first architecture uses hardware hash functions in order to reduces the overall simulation time, but lacks scalability. The second architecture uses an agent memory and a cell field memory. This improves the scalability and further increases the performance.