Skip to Main Content
Agent-Based Social Simulations have been largely used to study social phenomena. These kinds of simulations are related to reproduce real societies (e.g. social behaviors) using a bottom-up strategy. In classical social simulations, plausible macro-behaviors are only obtained if the social scientists input every features, components and relationships. Alternatively, modern agent-based modeling tools are being designed as discrete-event simulators. This architecture is monolithic, executes serially agent actions and have limited scalability. Aiming at the production of plausible and scalable social simulations, this research has two main goals: (i) introduction of a distributed architecture to build highly scale social simulations and (ii) help social scientists to modeling complex social phenomena in various granularities (i.e. level of detail). We used distributed computation concepts to balance the computational demand of a high scale social simulation. The proposed architecture allows simulations of large social region and their population. For validating the proposed approach and architecture, we simulated the consumption phenomenon on Recipe (the 4th-largest metropolitan area in Brazil). We performed a comparative analysis between the distributed computational approaches with an econometric model. Results revel significant gain in similarity when compared the data produced by the distributed model with the real phenomenon. The proposed architecture can be used for building support decision tools for training public officers on high impact social decisions.