Demand-Oriented Fog-RAN Slicing With Self-Adaptation Via Deep Reinforcement Learning

Abstract

As one of the key technologies for the 6G system, network slicing can enable the concurrent provisioning of heterogeneous quality of services (QoS) on various types of services under one Fog-RAN architecture. However, effective slicing of Fog-RAN is very challenging due to the diversification of QoS requirements and demand fluctuation. In this paper, we propose a demand-oriented two-tier Fog-RAN slicing trading framework to facilitate the slice generation between the mobile network operator (MNO) and service providers (SPs), and also the slice adaptation among SPs with the consideration of three service types, including data transmission, computation offloading and content caching. In Tier-I, a slice generation mechanism is developed to enable the MNO to provide customized slices for SPs by jointly scheduling 3D resources (communication, computing and caching resources) in a large time scale. In Tier-II, a slice adaptation mechanism is designed based on the deep reinforcement learning approach to facilitate SPs to perform effective resource adjustment on their own slices by purchasing/selling resources from/to other SPs in a small time scale. Numerical results show that the proposed scheme can satisfy the various QoS requirements through the 3D resource scheduling, and also improve the resource utilization owing to the proactive slice adaptation. Comparing with the traditional semi-dynamic slicing manner, the adaptive adjustment could improve the resource utilization about <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$30\%$</tex-math></inline-formula> .