On Enhancing Network Slicing Life-Cycle Through an AI-Native Orchestration Architecture.

Legacy experimental network infrastructures can still host innovative services through novel network slicing orchestration architectures. Network slicing orchestration architectures available in state-of-the-art have building blocks that structurally change depending on the problem they are trying to solve. In these orchestrators, life-cycle functions of network slices experience advances on numerous fronts, such as combinatorial methods and Artificial Intelligence (AI). However, many of the state-of-the-art slicing architectures are not AI-native, making heterogeneity and the coexistence and use of machine learning paradigms for network slicing orchestration hard. Also, using AI in a non-native way makes network slice management a challenger and shallow. Hence, this paper proposes and evaluates a distributed AI-native slicing orchestration architecture that delivers machine learning capabilities in all life cycles of a network slice. Carried experiments suggest lower error using distributed machine learning models to predict Radio Access Network (RAN) resource consumption in slicing deployed over different target domains.