Data Provenance for Multi-Agent Models in a Distributed Memory

Abstract

Multi-agent systems (MAS) assist with studying emergent collective behavior of individual entities in social, biological, economic, network, and physical systems. Applied to MAS, data provenance can support agent-based modeling by explaining individual agent behavior. However, there is no provenance support for MAS in a distributed setting. The Multi-Agent Spatial Simulation (MASS) library provides a framework for simulating agent-based models (ABM) at fine granularity, as multi-agent models (MAM), where agents and spatial data are shared application resources in a distributed memory. This Master’s thesis evaluates ProvMASS, a novel approach to capture MAM provenance in a distributed memory. Queries and performance measures indicate that an adaptive approach can generate provenance that explains coordination of distributed shared resources, simulation logic, and agent behavior while limiting performance overhead for long-running simulation (several hours) of large models (including hundreds of thousands of agents).