Oneshot Deep Reinforcement Learning Approach to Network Slicing for Autonomous IoT Systems

With the emergence of the Internet of Things (IoT) services, meeting multiple and diverse Quality of Service (QoS) requirements in networks has become a crucial issue. In the new 5G networks, network slicing is presented as the solution to provide a tailored QoS for different network services. This new technology offers better prospects for IoT services and applications. In fact, in modern IoT systems, the number of IoT devices increases, and these systems evolve to be autonomous IoT systems. QoS management must be done without human intervention, making conventional QoS management mechanisms unsuitable. In this article, we introduce an oneshot deep reinforcement learning (DRL) agent capable of autonomously receiving requests for slices and proposing a placement on the physical infrastructure that maximizes the total number of accepted requests while guaranteeing load balancing at the infrastructure resources level. By adopting a new paradigm located at the crossroads between the single DRL agent and the multiagent DRL, our agent manages to generate the placement decision of a slice request in one step, which makes it compatible with the European telecommunications standards institute (ETSI) standard. Numerous simulations and comparisons with six other algorithms allowed us to validate its effectiveness in real-time scenarios where learning from previous placements is required to improve future slice provisioning.