Towards Cognitive Autonomous Networks

Network Management Automation for 5G and Beyond

Stephen S. Mwanje, Christian Mannweiler, Stephen S. Mwanje, Christian Mannweiler

• Publisher: Wiley
• Year: 2020
• Print ISBN: 9781119586388
• eBook ISBN: 9781119586425
• Edition: 1
• Topic: Technology & Engineering
• Subtopic: Electrical Engineering & Telecommunications
• Index: Technology & Engineering

Learn about the latest in cognitive and autonomous network management Towards Cognitive Autonomous Networks: Network Management Automation for 5G and Beyond delivers a comprehensive understanding of the current state-of-the-art in cognitive and autonomous network operation. Authors Mwanje and Bell fully describe todays capabilities while explaining the future potential of these powerful technologies. This book advocates for autonomy in new 5G networks, arguing that the virtualization of network functions render autonomy an absolute necessity. Following that, the authors move on to comprehensively explain the background and history of large networks, and how we come to find ourselves in the place were in now. Towards Cognitive Autonomous Networks describes several novel techniques and applications of cognition and autonomy required for end-to-end cognition including: •Configuration of autonomous networks •Operation of autonomous networks •Optimization of autonomous networks •Self-healing autonomous networks The book concludes with an examination of the extensive challenges facing completely autonomous networks now and in the future.

1
The Need for Cognitive Autonomy in Communication Networks

Stephen S. Mwanje, Christian Mannweiler, and Henning Sanneck

Nokia Bell Labs, Munich, Germany

Communication networks have significantly evolved to the point that they have become very complex to operate. Concurrently, the demands placed thereon by the different stakeholders continue to increase. Users require more diversified, robust yet cheaper services; operators require network operation to be cheap and simple with short lead times to introduce new services while governments demand ubiquitous networks offering reliable services.

At the core of meeting all these demands is network automation – increasing the capabilities of networks to undertake more and more operational tasks which have historically been done manually. This already is an ongoing process, but as we motivate the next level of automation, we must look deep into the structure of networks to identify the areas of greatest promise for automation. This chapter takes this deep evaluation to set the baseline for the subsequent discussion. It seeks to answer these questions: (i) why do we need to pursue the path towards cognitive autonomy in networks? What is the gap between where we are now and the final destination of a Cognitive Autonomous Network (CAN)?

The chapter presents a high‐level overview of communication networks and the related Network Management (NM). It highlights the complexity of networks as justification for Network Management Automation (NMA) and discusses specific features of networks that are either the source or the embodiment of complexity in networks. Thereafter, it summarizes the NMA evolution and the existing framework within which NMA solutions have been developed. The discussion includes evaluation of the pressing NM challenges and the justification for evolving NM towards evermore cognitive capabilities. Finally, the chapter gives a taxonomy of the terminology used throughout the book and a preview of the contents presented in the remainder of the book.

Note that although the chapter and the rest of the book are heavily biased towards mobile networks, much of the discussion is generic and equally applies to other network domains. In fact, some chapters have sections that discuss concepts, implications and applications in both core and transport networks.

1.1 Complexity in Communication Networks

A communication network is responsible for getting data from one device (A) to another device (B), each of which may be a mobile device, for example, like a handheld device, a car, a robot or a fixed device like a fixed phone, a personal computer, or a server. In modern networks, the information which may include voice, simple text messages, files such as email/webpages or streaming media is all transmitted as packet data. The different network subparts contain devices and network elements (NEs) of varying size, whether considered in terms of physical size, form factor or available capacity. In the access' part, (where user terminals connect to the network), the typical network will have very many small‐sized network elements, i.e. devices with small form‐factor and where the maximum number of user terminals that can be concurrently served by the network element is also relatively small. The ‘core’ part of the network will instead have a small number of large‐size network elements which are capable of thousands of concurrently supported users or sessions.

The thousands of network elements that move data around the network create an extremely complex graph of devices, functions, and processes. The sources of this complexity are many and varied – ranging from the system and functional design of the individual nodes of the graph to the inter‐dependencies among these nodes, especially in the network's operational processes. Managing this complexity, which is a task delegated to network management, requires innovative approaches to be implemented in the network management systems for the networks to remain operable. The following sections discuss these sources of complexity and their implications for network management.

1.1.1 The Network as a Graph

A communication network may be represented as a graph [1]. It may, however, also be represented by a set of graphs each of which represents a distinct perspective of the network. In simple terms, a graph is an ordered pair G = (V, E) of a set V of vertices and a set E of edges between the vertices. The nature of relationships amongst the vertices and edges characterizes a graph as a simple graph, a multi‐graph, or a directed graph, etc. A communication network can be viewed as a graph on the specific view of the network. For example, on a global scale, a whole network can be a single graph if each transmitter and receiver point is considered a node/vertex with the communication links between them as the edges. Another graph, however, can be constructed for the set of neighbouring cells in a mobile network while the set of interfering cells would also be a different graph. These two graphs may also be considered as dynamic graphs if the user terminals are also considered as nodes on the graph, i.e. the terminals are dynamic leaves of the graph. It is, therefore, sensible to study networks in a way that each network is a set of graphs representing different perspectives.

Corresponding to the graph view of networks, network complexity directly translates into complexity of a graph, which may be measured in different perspectives that inherently signal the sources of the complexity. In general, there are three sources of complexity:

• Network protocol stack in each node: The abstraction of the physical system into a graph attempts to neglect the internal details of each node to focus on the interactions among nodes. However, each node can be complex with interactions amongst multiple graphs only realized within the node. For instance, individual nodes provide linkage between graphs of network routing and higher layer connected applications, e.g. for network transport dimensioning, network partitioning or for network caching.
• Scale: This, which is often the simplest dimension of complexity, is an indication of the number of nodes in the graph and the degree of interconnections among them. Correspondingly, it can be expressed in terms of the simplest measures for describing the structure of the graph, typically expressed in terms of the size or density.
• Connectivity: This is the most considered aspect of graphs – a measure of the degree of distribution/centralization and the homogeneity of its connectivity. By default, graphs assume a distributed system where value comes not from the individual nodes but from the complete unit that maximizes the connectivity amongst the nodes. As such, corresponding measures of complexity describe this connectivity in terms of the node or edge centrality measures, which describe the network positions of vertices and edges [2]. The most widely used measure is the edge betweenness, which identifies edges that are most crucial to maintaining a network's connectivity. However, there are many other measures including radius, closeness, etc. [3]

It is evident that these aspects of complexity are all present in networks and an understanding of this graph view is valuable to the process of designing NMA systems capable of addressing that complexity. NMA needs to leverage the multi‐graph of a network to address the NMA challenges. Of particular interest is the modelling of the graph view within NMA – either as a single graph representing a single model with a broad view or as a multi‐graph representing multiple different models with partial views that are related to each other. The subsequent sections describe, for the different sources of network complexity, the extent to which that complexity may be mapped to the complexity of a network graph.

1.1.2 Planes, Lay...

Table of contents

• Cover
• Table of Contents
• List of Contributors
• Foreword I
• Foreword II
• Preface
• 1 The Need for Cognitive Autonomy in Communication Networks
• 2 Evolution of Mobile Communication Networks
• 3 Self‐Organization in Pre‐5G Communication Networks
• 4 Modelling Cognitive Decision Making
• 5 Classic Artificial Intelligence: Tools for Autonomous Reasoning
• 6 Machine Learning: Tools for End‐to‐End Cognition
• 7 Cognitive Autonomy for Network Configuration
• 8 Cognitive Autonomy for Network‐Optimization
• 9 Cognitive Autonomy for Network Self‐Healing
• 10 Cognitive Autonomy in Cross‐Domain Network Analytics
• 11 System Aspects for Cognitive Autonomous Networks
• 12 Towards Actualizing Network Autonomy
• Index
• End User License Agreement