Joint Demand Forecasting and Network Slice Pricing for Profit Maximization in Network Slicing

Abstract

As a key technology of next-generation networks, network slicing enables networks to flexibly and efficiently fulfill the heterogeneous requirements of various services. In the slice-as-a-service business model, the service provider (SP) creates network slices to meet the service level agreements (SLAs) between the SP and users. In this article, we propose an SLA guaranteed network slicing framework (SLA-NS) to maximize the profit of the SP. In SLA-NS, we mainly focus on: i) network slice pricing; ii) resource demand forecasting; iii) on-demand resource allocation. For network slice pricing, we propose a two-layer game model to optimize the profit of the SP considering user prospects. The two-layer game model comprises a Stackelberg game between the SP and users and an evolutionary game among users. To achieve negligible SLA violations and low prediction error, we propose a resource demand predictor referred to as encoder-decoder long short-term memory with preference (LSTM-P). The on-demand resource allocation includes cross-slice resource preallocation and admission control based on prediction. The former reserves resources for active slices based on the predicted resource demands; the latter exploits LSTM-P to forecast the available resources in the long term to assist admission control. The simulation results show that the proposed SLA-NS yields at least 16.0% higher resource utilization and 19.5% higher SP profit than the benchmark allocation strategies.