Reliable and Efficient Agent-Based Modeling and Simulation

Agent-based models represent a primary methodology to untangle and study complex systems. Over the last decade, the need for more elaborate computing-demanding models gave rise to many frameworks and tools to run ABM simulations. Current state-of-the-art ABM tools either focus on ease of use, performance, or a trade-off between these two elements. Still, efficiency-oriented solutions (required for both large and small-scale simulations) are vulnerable to memory flaws which could invalidate the experiment results. This work aims to merge efficiency, reliability, and safeness under an innovative ABM software framework based on the Rust programming language. Our framework, krABMaga, is an open-source library that offers a high-level environment by exploiting metaprogramming and expandable visualization features. We equipped our library with a dynamic simulation monitoring system and model exploration and optimization capabilities over parallel, distributed, and cloud architectures. After having presented the overall architecture and functionalities of krABMaga, we discuss a performance comparison of our framework against the mostly adopted ABM software and the scalability potential of our simulation engine on a model calibration experiment running over an AWS EC2 virtual cluster machine. All code and examples models are available on GitHub.