The Matrix: An Agent-Based Modeling Framework for Data Intensive Simulations

Human decision-making is influenced by social, psychological, neurological, emotional, normative, and learning factors, as well as individual traits like age and education level. Social/cognitive computational models that incorporate these factors are increasingly used to study how humans make decisions. A result is that agent models, within agent-based modeling (ABM), are becoming more heavyweight, i. e. , are more computationally demanding, making scalability and at-scale simulations all the more difficult to achieve. To address these challenges, we have developed an ABM simulation framework that addresses data-intensive simulation at-scale. We describe system requirements and design, and demonstrate atscale simulation by modeling 3 million users (each as an individual agent), 13 million repositories, and 239 million user-repository interactions on GitHub. Simulations predict user interactions with GitHub repositories, which, to our knowledge, are the first simulations of this kind. Our simulations demonstrate a three-order of magnitude increase in the number of cognitive agents simultaneously interacting.

Authors

• Parantapa Bhattacharya University of Virginia
• Saliya Ekanayake Lawrence Berkeley National Laboratory
• Chris J. Kuhlman
• Christian Lebiere
• Don Morrison Carnegie Mellon University
• Samarth Swarup University of Virginia
• Mandy L. Wilson University of Virginia
• Mark G. Orr University of Virginia

Keywords

• agent-based simulation
• simulation framework
• distributed simulation