Adaptive Scheduling Algorithms for the Dynamic Distribution and Parallel Execution of Spatial Agent-Based Models

In previous work [7], we proposed a general framework for defining agentbased models (ABMs) and introduced two algorithms for the automatic parallelization of agent-based models: a general version P-ABMG for all ABMs definable in the framework and a more specific variant P-ABMS for “spatial ABMs”, which can utilize the additional spatial information to obtain performance improvements. Both algorithms can automatically distribute ABMs over multiple processors and dynamically adjust the degree of parallelization based on available computational resources throughout the simulation runs. However, they are not sensitive to inefficiencies in the sequence in which agents in each parallel simulation instance are updated. In this chapter, we introduce a minimal framework for describing ABMs and propose various asynchronous scheduling algorithms for agent-based simulations that address the update inefficencies of simulation schedulers. The proposed algorithms work in conjunction with P-ABMG and P-ABMS and allow for efficient simulation runs that can automatically and better utilize the asynchronous nature of parallel distributed agent-based simulations (including split-ups of specific simulation models and dynamic load-balancing). We demonstrate the significant performance gains of the proposed algorithms using an actual agent-based model used for studying female choice and foraging in biological research. Matthias Scheutz and Jack Harris Human-Robot Interaction Laboratory Cognitive Science Program and School of Informatics Indiana University Bloomington, IN 47406 e-mail: {mscheutz,jackharr}@indiana.edu