MASS CUDA: Abstracting Many Core Parallel Programming From Agent Based Modeling Frameworks.

摘要

Agent based modeling is the practice of simulating complex interactions by modeling the behavior of a single agent in the interaction, then observing the emergent behavior that occurs when many of those Agents interact. As simulation size increases, the computational requirements can become prohibitive, and parallelization of a simulation can be complex. As most users of Agent Based Models are not programmers by trade, most rely on frameworks to develop and parallelize processing of these models. NVIDIA's CUDA programming language is of interest as it can to harness the massively parallel capabilities of graphics processing units. This paper proposes an architecture for such a framework that uses the CUDA programming language to accelerate Agent Based Models while hiding the programming complexities from the end user.