Dynamic Resource Allocation for E2E Network Slicing with Both Public and Non-Public Slices

Yuxing Wang,Nan Liu,Zhiwen Pan,Xiaohu You
DOI: https://doi.org/10.1145/3670105.3670133
2024-01-01
Abstract:Network slicing technology provides a more flexible, intelligent, and efficient way to manage resources for various applications and business scenarios. However, dynamic resource allocation for slices to respond to changes in user demands is a challenging problem. This paper focuses on the dynamic resource allocation problem in the scenario where public and non-public network (NPN) slices coexist. We first formulate the above problem as a nonlinear integer optimization problem with an optimization objective of maximizing the operator’s profit. Then, in order to solve the above optimization problem, we propose a framework that combines multiple algorithms for resource allocation at different time scales. Firstly, the long short-term memory (LSTM) algorithm is used to predict the future resource demand of each slice and to assist in inter-slice resource allocation at large time scales. Next, the intra-slice resource allocation is transformed into a grouped knapsack (GK) problem and solved at each small time scale. Finally, we employ deep reinforcement learning (DRL), more specifically, the soft actor-critic (SAC) algorithm to solve the dynamic virtual network function (VNF) mapping problem at large time scales. The simulation results indicate that the proposed framework LSTM-GK-SAC exhibits excellent performance, not only ensuring the characteristics of the NPN slices but also adapting flexibly to the dynamic changes in the slicing environment.
What problem does this paper attempt to address?