Model-based reinforcement learning for router port queue configurations

Fifth-generation (5G) systems have brought about new challenges toward ensuring Quality of Service (QoS) in differentiated services. This includes low latency applications, scalable machine-to-machine communication, and enhanced mobile broadband connectivity. In order to satisfy these requirements, the concept network slicing has been introduced generate slices with specific characteristics. meet requirements slices, routers switches must be effectively configured provide priority queue provisioning, resource contention management adaptation. Configuring from vendors, such as Ericsson, Cisco, Juniper, traditionally an expert-driven process static rules for individual flows, which are prone sub optimal configurations varying traffic conditions. this paper, we model internal ingress egress queues within via a queuing model. The effects changing configuration respect priority, weights, flow limits, packet drops studied detail. is used train model-based Reinforcement Learning (RL) algorithm policies prioritization, fairness, congestion control. efficacy RL policy output demonstrated over scenarios involving policing, shaping, one-hop router coordinated conditioning. evaluated real application use case, wherein statically proved desired QoS requirements. Such automated will critical multiple 5G deployments patterns.