A Distributed, Multi-Agent Approach to Reactive Network Resilience

Critical network infrastructures are communication networks whose disruption can create a severe impact on other systems. Multi-agent systems have been successfully used to protect critical network infrastructures, with approaches ranging from reasoning about secure design and policies to multi-agent intrusion detection systems (IDS). However, there is little research on the possibilities for multiagent systems to react to known intrusions. In this paper we propose a multi-agent framework for reactive network resilience, that is, to allow a network to reconfigure itself in the event of a security incident so that the risk of further damage is mitigated. The proposed framework takes advantage of a risk model based on multilayer networks and of distributed belief propagation techniques to agree on a new, more resilient configuration of the network in the event of an attack. We compare our proposal with a number of centralized optimization and multi-agent negotiation techniques. Experiments show that our proposal outperforms the reference approaches both in terms of risk mitigation and performance