Network Intelligence for NFV Scaling in Closed-Loop Architectures

Characteristics like self-managing, self-adaptation, and self-organization are the main objectives of intelligent network operation. AI and Machine Learning (ML) algorithms will enable future networks to operate entirely autonomously. However, current network architectures are not fully prepared to include and properly handle the promised Network Intelligence (NI). This article looks at scaling, one key Management and Orchestration (MANO) operation that allows the network to adapt to unexpected changes. We show how different scaling methods can fit the Monitor-Analyze-Plan-Execute over a shared Knowledge (MAPE-K), a well-established framework that allows architectural-based adaptation. Using a cloud-based scenario, we compare and highlight architectural differences between two prominent scaling methods, one based on Reinforcement Learning (RL) and the other based on classical control theory, showing that only the data-driven approach is adaptable enough to achieve automation. We conclude the article by pointing toward future research in autonomous, adaptive networks.