Resource Reservation within Sliced 5G Networks: A Cost-Reduction Strategy for Service Providers

Abstract

In future cellular networks, licensed Mobile Network Operators (MNOs) will be able to dynamically allocate dedicated resources to third party service providers (SPs) via an open market initiative known as network slicing. In this paper, we investigate the problem of resource reservation, whereby the SP must reserve future resources (slices) from the MNO in order to guarantee a minimum quality of service (QoS) for its future traffic demands. However, the reservation of these resources will come at an additional cost to the SP, which motivates the main objective of this paper: To develop a strategy that will allow the SP to determine the optimal amount of resources it should reserve ahead of time so that its future QoS constraints are met with minimal risk of revenue loss (i.e. from over reservation). To achieve our objective, we formulate this problem into a time series prediction task and employ a data-driven approach using two state-of-the-art machine learning concepts, namely deep neural networks (DNNs) and long short-term memory (LSTM) recurrent networks, to minimise the amounts of over/under reservations within a short-time prediction context. We find that our solutions demonstrate high performance against a real-world data set containing cellular data bytes for the city of Shanghai. In addition, we demonstrate the robustness of our designs by comparing their performance against a baseline solution based on the popular ARIMA forecasting model.