An Overview of Machine Learning-Enabled Network Softwarization for the Internet of Things

TL;DR

The paper tackles the challenges of IoT scalability and heterogeneity by advocating ML-enabled network softwarization, integrating SDN and NFV to create programmable, adaptable IoT networks. It surveys IoT fundamentals, SDN/NFV architectures, and ML taxonomies (SL, UL, RL, FL), then reviews recent ML-enabled network softwarization efforts, including RL/DRL for SFC/VNF-FG placement and edge FL. The authors highlight gaps like underutilization of NFV in ML contexts, energy-awareness, and a dearth of standardized datasets, offering future directions such as edge computing, distributed multi-controller softwarization, and stronger security focus. The work serves as a roadmap for researchers and practitioners to build scalable, energy-efficient, secure, and intelligent IoT networks through integrated ML-enabled softwarization.

Abstract

The Internet of Things (IoT) has evolved from a novel technology to an integral part of our everyday lives. It encompasses a multitude of heterogeneous devices that collect valuable data through various sensors. The sheer volume of these interconnected devices poses significant challenges as IoT provides complex network services with diverse requirements on a shared infrastructure. Network softwarization could help address these issues as it has emerged as a paradigm that enhances traditional networking by decoupling hardware from software and leveraging enabling technologies such as Software Defined Networking (SDN) and Network Function Virtualization (NFV). In networking, Machine Learning (ML) has demonstrated impressive results across multiple domains. By smoothly integrating with network softwarization, ML plays a pivotal role in building efficient and intelligent IoT networks. This paper explores the fundamentals of IoT, network softwarization, and ML, while reviewing the latest advances in ML-enabled network softwarization for IoT.