Multi-User Service Throughput Estimation for Adaptive Resource Management in 5G-Advanced and Beyond Sliced Networks

Due to the random arrival of traffic in 5G-Advanced networks, optimum slice resource management has explicitly become challenging with an increase in the number of users. Each network slice is designed to map its specific characteristics, such as service demands and bandwidth to performance metrics, such as throughput and delay. These slice-level metrics are continuously monitored during deployment to check whether the resources have to be reconfigured to meet the service level agreements (SLAs) while minimizing resource over-provisioning. Thus, developing accurate slice-level performance models has gained interest for cellular network operators. Most of the prior work has relied on pre-negotiated operator-defined SLAs with a focus on fractional resource allocation. While there has been limited work on slice throughput estimation, the effect of multiple access within a slice due to competitive behaviour among users is a complex open research problem. This paper presents a novel comprehensive analysis tackling User Service Throughput Estimation (USTE) within and across multiple slices. USTE works in two stages. In the first stage, the channel distributions of users hosting services are modeled, followed by dynamic conditional throughput estimation in the second stage. For new users requesting a slice, the throughput is re-estimated and re-configured across prior users hosting services with the same slice. The proposed method will facilitate the system to detect resource over-provisioning, slice capacity problems and SLA degradations through network dimensioning and re-dimensioning. Analytical and simulation results show that USTE performs scalable user scheduling yielding 55% throughput gains compared to the state-of-the-art.