Thesis title: Machine Learning for autonomous network control in next generation backbone networks
In this Thesis we investigated the exploitation of the potential of Machine Learning algorithms in the context of networking problems. The main objective is to propose solutions strengthened by artificial intelligence algorithms to improve the performance of current network policies. The comparison is based on the evaluation metrics related to the quality of the service provided. Two main scenarios are investigated in the work. In the first, we propose the L-DiffServ framework that can be integrated into the DiffServ protocol, which provides to the network operator the ability to dynamically increase the granularity of the classification and at the same time to further differentiate a traffic that does not present service differentiation. In the second scenario we propose a solution referred as iLLC, which exploiting Reinforcement Learning algorithms, keeps under control the local links utilization respect to the node where the agent is installed. The network scenario considered is a segment routing network which provides the flexibility to redefine a rerouting operation at node level. The proposal starts with the implementation of a single node agent in the network, and then move to a general environment with multiple agent nodes. The improvements respect to the current situation, without enabling the iLLC, is shown in terms of utilization and end to end delay.