Active Queue Management in Disaggregated 5G and Beyond Cellular Networks Using Machine Learning

In the context of 5G and beyond cellular networks, this paper delves into Active Queue Management (AQM) implementation in high-latency environments within disaggregated Radio Access Network (RAN) deployments, addressing bufferbloat while improving overall end-to-end network performance. While researchers investigate AQM algorithms to mitigate bufferbloat in monolithic RAN deployments, they overlook the inherent capabilities of disaggregation in 5G and beyond cellular networks, which involves intricate cross-layer communication among distinct network entities housing different protocol stack layers. Our study explores the 5G architecture, investigates previous research on AQM, identifies challenges arising from these algorithms in dis aggregated network configurations, and proposes a comprehensive scheme for managing AQM in these configurations. To facilitate our approach, we leverage RAN Intelligent Controller (RIC) entities equipped with Artificial Intelligence (AI) and Machine Learning (ML). We evaluated our novel solution by implementing a previously proposed AQM algorithm, known as Dynamic RLC Queue Limit (DRQL), in a disaggregated RAN deployment within a high-latency network environment and assessing its effectiveness through the Quality of Service (QoS) achieved at the NITOS testbed, utilizing OpenAirInterface5G (OAI5G).