SNAF: DRL-Based Interdependent E2E Resource Slicing Scheme for a Virtualized Network

Abstract

The fifth generation (5G) mobile cellular network relies on network slicing (NS) to satisfy the diverse quality of service (QoS) requirements of various service providers operating on a standard shared infrastructure. However, the synchronization of radio access network (RAN) and core network (CN) slicing has not been well-studied as an interdependent resource allocation problem. This work proposes a novel slice-to-node access factor (SNAF)-based end-to-end (E2E) slice resource provisioning scheme and deep reinforcement learning (DRL)-based real-time resource allocation algorithm for E2E interdependent resource slicing and allocation, respectively, specifically for RAN and CN. To ensure effective resource slicing and allocation, we consider the versatile user equipment (UEs) QoS requirements on transmission delay and data rate. Notably, the SNAF-based scheme provides proper resource provisioning and traffic synchronization, while the DRL-based algorithm allocates radio resources based on affordable traffic and backhaul resources. Based on the 5G air interface, we conduct system-level simulations to evaluate the performance of our proposed methods from various perspectives. Simulation results confirm that our proposed SNAF and DRL-based interdependent E2E resource slicing and allocation techniques achieve better E2E traffic-resource synchronization, and improve the QoS satisfaction with minimal resource utilization compared to other existing benchmark schemes.