CyberGIS-ABM: Scaling Complex Spatial Simulations to HPC

Keywords: HPC, CyberGIS, ABM
Abstract Type: Paper Abstract

Abstract

Recent and ongoing crises underscore the complexity surrounding interrelationships between human actions and impacts such as disease spread and evacuation congestion. Growing recognition of the far reaching impacts made by individual actions calls for approaches that can capture large-scale processes that depend on small scale actions and interactions.

A powerful tool for modeling complex processes can be found in agent-based models (ABMs) through "bottom-up" reasoning. In an agent-based model, actions are encoded for individual actors, or agents. Agents may take actions and influence the environment and other agents. However, scaling up an agent-based model requires more significant computational resources, especially when the model is based on spatial network representations.

To address these scaling limitations, we have developed a CyberGIS-ABM framework for synergizing high-performance computing and network science for scalable agent-based modeling of human-environment interactions. This framework is designed to directly harness high-performance computing resources using Message Passing Interface (MPI) and object-oriented programming.

This presentation will illustrate approaches for efficiently scaling ABMs to effectively utilize high-performance computing resources and demonstrate scaling efforts and computational results on Frontera, a large-scale computing cluster and discuss avenues for further spatially-targeted optimization. This work aims to extend the possibilities for informed analyses in the context of large-scale emergency evacuation modeling.