Reasoning in agent-based network management

An increasing complexity of mobile networks and use cases is reflected in requirements for network management. Advanced automation is required to address this challenge in an economically feasible manner. We describe a system based on agent mapping as a platform for automation for network management in 5G networks and beyond. We use classical and probabilistic reasoning for composing solutions to complex requests by means of relatively simple software agents. We describe different variants of the approach in terms of capabilities, ranging from triple store only to system with semantic and probabilistic reasoning functionalities. This approach provides flexibility for network functionality evolution and facilitates software reuse and is compatible with the use of task-specific machine learning algorithms in network management agents. We describe test system used for evaluating the concept, as well as use case evaluation obtained with it.