SWARM Simulation of Multi-Agent Fault Mitigation in Large-Scale, Real-Time Embedded Systems

This paper presents a SWARM multi-agent simulation of fault mitigation within BTeV, a large-scale, realtime embedded system. The Real-Time Embedded Systems (RTES) group collaborates on designing real-time embedded intelligent software to ensure data integrity and fault tolerance within BTeV, a triggering and data acquisition system for particle-acceleratorbased High Energy Physics experiments at Fermi National Laboratory. The hardware layout spans 2,500 digital signal processors and approximately 2,500 Linux computers. Adaptive and small in footprint, very lightweight agents were designed to apply both reactive and proactive rules to accomplish fault tolerance within the system. The scale and real-time requirements of this system make it ineffective to design a traditional expert system that relies on centralized processing to determine appropriate rule actions for every possible system state. Instead, a decentralized approach based on Rodney Brooks’ subsumption architecture is presented. A SWARM simulation is used to investigate emergent behavior of the multi-layered, distributed approach to fault mitigation within the RTES/BTeV environment.