A Development of Dynamic End-To-End Slicing Network Along with the Offline SLA-Method Integrated with Deep Learning for 5G/6G Networks

Using SDN, or software-defined networking, and Network Function Virtualization, or (NFV), we address the problem of resource provisioning in fifth-generation (5G) networks to enable end-to-end dynamic slicing. The core of our method is to dynamically assign separate portions of Physical Resource Bricks (PRBs), basic processing resources, transport capacity, and components like The information Forwarding Elements (DFE) plus network connections to tenants, logical operators of different slices. We use large datasets of key performance indicators (also known as KPIs) taken from a live mobile phone network fitted with traffic sensors in order to create a forecast model for the traffic of slices. We offer a technique that makes use of a soft-gated Randomised Unit (GRU) based low-complexity slices traffic prediction. Furthermore, we build joint multi-slice DNN s (deep neural networks) for each virtual network role and train them to predict the needed resources, guaranteeing adherence to two crucial service level contracts (SLAs): resource bounds-based SLA and violation rate-based SLA. We present dataset-dependent generalised non-convex constraints as a solution to the difficulty of optimisation challenges related to DNNs. The offline optimisation procedure of the DNNs, which is solved using a non-zero summing two-player game strategy, smoothly incorporates these limitations. Throughout our methodology, we stress the importance of underlying hyperparameters that in striking a balance between the trade-off between overstocking and preserving slices' isolation. Lastly, we do a thorough closed-form analysis including reliability theory in order to determine the bottom limit of the reliable convergent probability. In addition, we investigate how the violation rate affects this convergence probability, offering important information on the dependability of our suggested system.