Automating Network Management With Artificial Intelligence: In Software Networks and Beyond

Introduction

Mobile communications are one of the areas that has grown tremendously in the Information and Communication Technologies (ICT) market. It is expected that trillions of various devices, such as smartphones and tablets, medical devices, traffic and security cameras, etc., will connect to cellular networks by 2020. As a result, future mobile networks need to support the huge amount of resources that will be consumed, which will lead to a dramatic increase of network traffic (Cisco, 2016). Meanwhile, the emergence of new types of devices enables various new applications that affect different spheres of human life. Likewise, the ongoing deployment of the Internet of Things will foster the development of machine-to-machine communications, which complement the dominating human-centric communications of today (2018).

This would subsequently lead to a diversity of communication characteristics. Both trends will raise new requirements on network scalability, data rates, latency and reliability. The forecasted huge increase in demand for network capacity, and strict and diverse requirements raised from new communication patterns may not be adequately addressed along with the evolution of existing technologies. Therefore, research on new technologies is necessary in order to complement current ones. Additionally, techniques for the virtualization of network functions are becoming more and more mature. Such techniques will undoubtedly play an important role in the new generation of networks: for example, 5G promoters are investigating how virtualization and orchestration capabilities can be used to dynamically accommodate ever-changing resource demands.

5G networks will also rely upon other technologies, such as network densification and infrastructure sharing, to address the challenges and requirements faced by today's wireless networks (including cellular networks). It is not hard to foretell that the complexity of network management will become one of the biggest challenges to be addressed by 5G infrastructures, because of the conglomeration of technologies. To cope with similar challenges in 3G and 4G networks, self-administering and self-managing networks have been extensively researched, as presented later in the chapter. The fundamental work in the area of automation or advanced and smart automation is known as autonomic computing and autonomic networking where self-x (e.g., self-configuring, self-healing) functions were defined to ease, simplify and automate network operations since 2001 by IBM Manifesto (Kephart, 2003).

Both the come-back of Artificial Intelligence (AI), in particular Machine Learning (ML) and deep learning owing to the increasing computation power and the transformation of networks towards more softwarization, attracted much academic and industrial attention to cognitive management of networks.

This chapter primarily discusses the effort conducted by the European collaborative project CogNet (CogNitive Networks): Towards automation with Artificial Intelligence, a cognitive management framework and its inner modules are presented. The project aims at making a major contribution towards the automation of network and infrastructures management by investigating existing and devising novel ML algorithms tuned for available network data in order to yield insights, detect meaningful events and conditions, and respond correctly to them.

The chapter is structured as follows. The background section presents the autonomic and cognitive management, policy-based management, the concept of big data as well as an overview of current projects related to automated network management. Then, the authors present a cognitive management framework aiming to realize an automated control loop for the management operations with Artificial Intelligence. To exemplify this framework two use cases namely, Urban Mobility Awareness in 3G/4G networks and SLA Enforcement in an NFV (Network Function Virtualization) context are detailed. For both use cases, the workflows are presented to show how the framework building blocks interact together. In addition, a focus on the data and algorithms shows how to automate network management with machine learning and deep learning techniques.