A Parallel Computing Approach for Spatially-Explicit Agent-Based Models

Eric Shook, Wenwu Tang, and Shaowen Wang
University of Illinois at Urbana-Champaign

Spatially-explicit agent-based models are recognized as an effective computational framework for the investigation of dynamic geographic phenomena. However, large-scale agent-based models often pose significant computational challenges that need to be addressed through the use of high performance computing environments. This paper identifies a series of interprocessor communication issues that are important to the development of parallel spatially-explicit agent-based models. With consideration of these issues, a parallel computing approach based on a distributed communication strategy is presented. The communication strategy is implemented to exploit high performance computing environments with hundreds of teraflops peak performance in a scalable way. We examine the performance cost of the identified computation and communication issues on the National Science Foundation TeraGrid. In particular, we focus on investigating the influence that spatial characteristics of agents and environments have on the computational performance of agent-based models. Experimental results show that our parallel computing approach is well-suited to large-scale agent-based modeling of dynamic geographic phenomena.

Key words: Agent-based model, geo-simulation, high performance computing, parallel processing, performance evaluation

– Eric Shook Oct 16, 5:24 (Note: I also removed the reference to Sugarscape)

Return to scalableabm