Enabling Architecture for Distributed Intelligent Network Softwarization for the Internet of Things
Mohamed Ali Zormati, Hicham Lakhlef
TL;DR
The paper addresses the challenge of managing massively heterogeneous and resource-constrained IoT deployments. It proposes a distributed intelligent network softwarization architecture that integrates SDN, NFV, and FL across a five-layer design, including a global root controller as FL aggregator and multiple domain controllers as learning nodes. This setup aims to improve scalability, reliability, and privacy by performing distributed learning without sharing raw data and by enabling autonomous decision-making with DRL potential. The work provides a blueprint for future evaluations of QoS, energy efficiency, and IoT-specific metrics, signaling practical impact for next-generation IoT networks.
Abstract
The Internet of Things (IoT) is becoming a part of everyday life through its various sensing devices that collect valuable information. The huge number of interconnected heterogeneous IoT devices poses immense challenges, and network softwarization techniques are an adequate solution to these concerns. Software Defined Networking (SDN) and Network Function Virtualization (NFV) are two key softwarization techniques that enable the realization of efficient, agile IoT networks, especially when combined with Machine Learning (ML), mainly Federated Learning (FL). Unfortunately, existing solutions do not take advantage of such a combination to strengthen IoT networks in terms of efficiency and scalability. In this paper, we propose a novel architecture to achieve distributed intelligent network softwarization for IoT, in which SDN, NFV, and ML combine forces to enhance IoT constrained networks.