Towards a Generalized Reconfigurable Agent-Based Architecture: Stock Market Simulation Acceleration

Agent-based simulations have been used in domains ranging from biology to economics. In order to more accurately simulate systems, agent based simulations are becoming larger and more complex. Simulations could benefit from the speed and scale provided by a generalized agent-based architecture. In order to understand which features are relevant for such an architecture, we create a scalable implementation of an agent-based stock market model and compare it to an existing market simulation. The scalable stock market simulation uses several distributed markets rather than a single centralized market. A hardware implementation with support for reconfigurable agent logic and various market topologies is described. Synthesis results for simulations with up to 100 agents are presented. Lessons learned form this domain specific implementation will be applied in future work to a generalized agent-based architecture.