NWDAF-aided resource orchestration in 5G systems under QoS constraints

Nowadays, mobile networks are designed to provide very heterogeneous connectivity services, addressed to numerous kinds of different customers and vertical sectors. Such services come with varying performance requirements in terms of latencies, quality of service, reliability, etc. To satisfactorily meet these requirements and simultaneously utilize resources efficiently, the network needs to integrate advanced flexible and automated management and orchestration systems. In this work, we focus on the operations of an external orchestrator that interacts with the 5G core network to leverage its exposed resource allocation and configuration mechanisms. Namely, acting as an application function, the orchestrator consumes the services of the Network Exposure Function (NEF) to: i) collect metrics on the status of the network components and performance indicators on the data traffic from the Network Data Analytics Function (NWDAF), and ii) whenever needed, based on the gathered data, take action and re-allocate the network resources to guarantee that the required quality of service is delivered. These mechanisms can be leveraged, for instance, in use cases where external service providers pay the network operator a proportional amount to the consumed resources, and therefore, they want to optimally and efficiently control the trade-off between exploited resources and user's quality of experience. In this thesis, we offer a solution to the described management and orchestration problems based on the commercial HPE 5G core network. We developed NWDAF and NEF prototypes that are integrated with the HPE software and that implement the needed services to deploy a closed resource control and orchestration loop. To test and validate the proposed approach, we devised a reactive and a proactive resource re-allocation strategy, which address different quality of service requirements. The proactive strategy is based on an AutoRegressive Integrated Moving Average (ARIMA) model, which predicts the load (in terms of throughput) of the 5G User Plane Functions (UPFs). It allows to reduce the waste of resources related to the activity time of a secondary UPF by 65.79% and 53.09% in two different targeted QoS scenarios, compared to a baseline naive approach, guaranteeing good performance in terms of delivered QoS. The reactive strategy, instead, is less conservative and, at the price of slightly lowering the offered QoS, yields a higher reduction of the secondary UPF activity time, of 68.02% and 54.06% compared to the baseline approach in the same QoS scenarios.