Distributed Search Efficiency and Robustness in Service oriented Multi-agent Networks

We study decentralized searches in a large service-oriented agent network and investigate the influences of multiple factors on search efficiency. In this study we focus on overall system robustness and examine search performance in unstable environments where individual agents may fail or a system-wide attack may occur. Experimental results show that searches continue to be efficient when a large number of service agents become unavailable. Surprisingly, overall system performance in terms of a search path length metric improves with an increasing number of unavailable agents. Service unavailability also has an impact on the load balance of service agents. We plan to conduct further research to verify observed patterns and to understand related implications on system architecture design.