Replication-Based Self-healing of Mobile Agents Exploring Complex Networks.

Multi-agent based approaches can offer highly scalable, robust and flexibleways to provide data-collection and synchronisation services in large-scale dynamic distributed environments, ranging from physical terrains to sensor networks, computing Clouds, and the Internet of Things (IoT). The network topology of the targeted distributed system, as well as the agents' exploration algorithm, have an important impact on service performance and consequently on robustness in case of failure-prone agents. In previous works we have proposed a pheromone-based agent exploration algorithm that performs best in most targeted environments. We have also identified the network topology characteristics that are most sensitive to agent failure. In this paper, we propose a replication-based selfhealing approach that enables agents to complete a data-synchronisation task even for high-failure rates, in failure-sensitive network topologies. System nodes can learn and estimate time-outs dynamically, so as tominimise false positives. We evaluate overheads incurred by agent replication, in terms ofmemory consumption and message communication. The reported findings can help design viable multi-agent solutions for a wide variety of data-intensive distributed systems.