Integrate: Intent-Driven Router Queue Depth Configuration via Reinforcement Learning

With the emergence of 6G networks, autonomous configuration management of transport routers and switches becomes crucial. Active Queue Management (AQM) policies in routers and switches are employed to effectively maintain network slice Quality of Service (QoS) in cellular networks. However, routers and switches are statically configured with specific buffer (queue depth) capacities that can restrict performance, especially within intent-driven networks. In this paper, we propose an Artificial Intelligence (AI) based approach towards intelligent management of queue depths in 6G transport networks. Through the use of exhaustive experiments on a network digital twin, we characterize the effect of changing queue depth on throughput, packet drop and packet delay of various network flows. The experimental data is then used to train a model based reinforcement learning (RL) model to learn and optimally configure the router queue depth. Changes in intents can be dynamically mapped to the reward structure used to execute the RL policies. This technique is generic and may be applied to multiple types of routers and switches at cell site, front haul, back haul and data-center fabrics.