Scalable Multi-agent Simulation Based on MapReduce

Abstract Jason is perhaps the most advanced multi-agent programming language based on AgentSpeak. Unfortunately, its current Java-based implementation does not scale up and is seriously limited for simulating systems of hundreds of thousands of agents. We are presenting a scalable simulation platform for running huge numbers of agents in a Jason style simulation framework. Our idea is (1) to identify independent parts of the simulation in order to parallelize as much as possible, and (2) to use and apply existing technology for parallel processing of large datasets (e. g. MapReduce ). We evaluate our approach on an early benchmark and show that it scales up linearly (in the number of agents).

Authors

• Tobias Ahlbrecht
• Jürgen Dix
• Niklas Fiekas

Keywords

No keywords are indexed for this paper.