Network slicing is a technique introduced by 3GPP to enable multi-tenant operation in 5 G systems. However, the support of slicing at the air interface requires not only efficient optimization algorithms operating in real time but also its tight integration into the 5 G control plane. In this paper, we first present a priority-based mechanism enabling defined performance isolation among slices competing for resources. Then, to speed up the resource arbitration process, we propose and compare several supervised machine learning (ML) techniques. We show how to embed the proposed approach into the ITU-T standardized ML architecture. The proposed ML enhancement is evaluated under realistic traffic conditions with respect to the performance criteria defined by GSMA while explicitly accounting for 5 G millimeter wave channel conditions. Our results show that ML techniques are able to provide suitable approximations for the resource allocation process ensuring slice performance isolation, efficient resource use, and fairness. Among the considered algorithms, polynomial regressions show the best results outperforming the exact solution algorithm by 5–6 orders of magnitude in terms of execution time and both neural network and random forest algorithms in terms of accuracy (by 20–40 %), sensitiveness to workload variations and training sample size. Finally, ML algorithms are generally prone to service level agreements (SLA) violation under high load and time-varying channel conditions, implying that an SLA enforcement system is needed in ITU-T's 5 G ML framework.

Network slicing is a promising technique to enable multi-tenant operation in fifth generation (5G) cellular systems. However, efficient implementation of this technique at the air interface requires an adequate resource allocation policy that provides slice isolation and takes into account the heterogeneity of services and their performance requirements. We propose a new slicing scheme aimed at slice performance isolation, efficient capacity utilization, and fair resource allocation among all users. The policy ensures a slice-specific minimum data rate to all slice users as long as their number remains within a contracted limit, yet provides higher data rates whenever capacity is available. Specifically, we incorporate the best features of the complete partitioning and complete sharing policies, while ensuring flexibility and customization for network operators. The proposed scheme is based on an iterative solution of a convex programming problem characterized by polynomial complexity, which makes it suitable for on-line implementation. Associated algorithms are also proposed. Our numerical results demonstrate that the proposed slicing scheme is capable to accommodate heterogeneous traffic and provides performance isolation of slices while maintaining resource utilization at the level of complete sharing. Depending on the system parameters, the proposed scheme allows to improve session loss probability by an order of magnitude as compared to static slicing, while keeping the user data rates comparable to that of the complete sharing scheme and improving the average user satisfaction index by up to 90%. Overall, the proposed scheme results in efficient resource utilization compared to slicing policies in which slice isolation is provided via reservation by setting fixed upper or lower capacity bounds.