MARIO: Multi-Agent ResIlience framewOrk for Network Slicing in Tactical Networks

Abstract

Network slicing constitutes a paradigm shift as it transforms a 5G network into a set of versatile sub-networks for designated users, with specific requirements on security levels and quality of service (QoS) demands. Therefore, organizations with very high security and non-negotiable QoS requirements, such as the military, are leveraging the utilization of 5G network slicing for their operations. However, some network management challenges remain before constructing a resilient network with 99.999% reliability to different attacks, while providing isolation, high throughput and low latency. In this respect, we propose a reinforcement learning-based multi-agent resilience framework, which comprises centralized training using global information and decentralized decision-making by individual agents, each corresponding to an access point, to autonomously adapt network slicing configurations based on the evolving tactical landscape. The proposed framework continuously assesses network conditions and threat scenarios, to dynamically allocate the required resources and mitigate vulnerabilities. Numerical results exhibit that our proposed framework effectively defends against different adversarial actions and maintains operational continuity without compromising the QoS.