Abstract
Network Slicing is being recognized as one of the most promising techniques to improve QoS in 5G networks. It enables physical network sharing between isolated virtual slices. In this context, to meet specific QoS requirements for each slice, physical network resources should be efficiently reserved and channels should be optimally assigned for each slice member. In this paper, we evaluate in-depth various slicing strategies in large scale realistic industry 4.0 scenarios. More precisely, we compare a static slicing strategy to dynamic estimation and prediction-based algorithms. The former considers Maximum Likelihood Estimation (MLE) and the latter is based on Mini-batch Gradient Descent (MBGD) algorithm. The inter slicing resource methods are evaluated over NS3 simulator over which an extensive study is performed. We evaluate the impact of gateway (GW) positioning inside the industry, applications periodicity, parameter configuration, and load. Results show that the prediction-based strategy outperformed the estimation-based slicing algorithm in terms of respecting delay thresholds, decreasing energy consumption, and increasing the battery lifetime of IoT devices.