Implementation of Massive Agent Model Using Repast HPC and GPU

Abstract

Multi-Agent Simulation (MAS) is efficient for analysis of various social mechanisms. Recently, there are many studies on massive agent model to explain more complex social phenomena. Then, we aim for implementation of large scale simulation model using Repast HPC toolkit, a platform for massive agent model. In this article, we build "Schelling Segregation Model" for spatial model using geospatial data provided OpenStreetMap, an open source project creating a free editable map. In this model, agents are located continuous space , not grid in original. When an agent is "unhappy" and migrate to new location, it costs agents some simulation time depending on distance between old location and new one. This article reports simulation results using Japanese cities and verification result about execution time.