GPU Delegation: Toward a Generic Approach for Developping MABS using GPU Programming

Abstract

Using Multi-Agent Based Simulation (MABS), computing resources requirements often limit the extent to which a model could be experimented. As the number of agents and the size of the environment are constantly growing in these simulations, using General-Purpose Computing on Graphics Units (GPGPU) appears to be very promising as it allows to use the massively parallel architecture of the GPU (Graphics Processing Unit) to do High Performance Computing (HPC). Considering the use of GPGPU for developing MABS, the conclusions of Perumalla and Aaby's work \cite{Perumalla2008} in 2008 was twofold: (1) data parallel execution capabilities of GPU can be used effectively in MABS and afford excellent speedup on models and (2) effective use of data parallel execution requires resolution of modeling and execution challenges at the cost of a decrease in modularity, ease of programmability and reusability. In this paper, we propose to study through experiments if the conclusions and issues outlined by Perumalla and Aaby are still true despite the evolution of GPGPU and MABS. To this end, we use the GPU environmental delegation principle on four models in order to compare CPU and GPU implementations. Then, we discuss and analyze the results from both a conceptual and a performance point of view.