LARGE-SCALE DISTRIBUTED AGENT-BASED SIMULATION FOR SHOPPING MALL AND PERFORMANCE IMPROVEMENT WITH SHADOW AGENT PROJECTION

In this paper, we introduce the agent-based simulation of a shopping mall with walking and purchasing behavior model and consider the performance of distributed parallel execution. To utilize the agent-based simulation for decision support, distributed parallel execution of large-scale agent-based social simulations is important for evaluating the complex behavior of a realistic number of people with acceptable performance. For this purpose, today's agent-based simulation frameworks often provide the functionality to transfer agents from one node to another. However, intelligent social agents tend to contain a large amount of data including demographics, preferences, and history. Hence, the transfer of such an agent incurs a heavy communication cost that has an adverse effect on performance. To improve the performance of distributed agent-based simulation, we introduce a shadow agent that is a lightweight entity projected among nodes with only required information such as the position and speed required to calculate interaction between agents.