Network Slice Reuse in 5G Network: A Machine Learning Approach

The rise of 5G networks has been fundamental in these last decades, giving the possibility of handling the user demand in a sustainable way. The concept of a flexible and standard architecture brought by the 5G-TRANSFORMER project and later by the 5Growth project has revolutionized the approach in 5G networks. The network slicing approach is the key factor of this thesis. As a matter of fact, it enables a logical and physical separation in terms of amount of resources, with the possibility of sharing and scaling them. Further improvements can be done in these terms of optimization of resources, essential in handling a wide range of services with different requirements. Our work is focused on automatizing the resources optimization by the use of a machine learning model. This model should return to the system the best configuration parameters for the context analyzed, resulting in a proper resource utilization. In order to achieve this goal, we need to define and create a dataset relative to a network context which will be obtained through extensive service instantiation requests simulations, considering all the possible configuration parameters decided in advance. Different machine learning classification algorithm will be considered and the best in terms of accuracy will be chosen. The model is then given to the system through a new architectural layer defined by 5Growth, the AIML platform, responsible for handling model request and creation with respect to the data monitored and the vertical requirements. The final step of our work is to implement the model in the reality with different services as the context requires, simulating instantiation requests using the correct parameters.