AI-Based Advanced Optimization Techniques for Edge Computing E-Book

• Herausgeber: John Wiley & Sons
• Kategorie: Wissenschaft und neue Technologien
• Sprache: Englisch

Seitenzahl: 680

Veröffentlichungsjahr: 2025

Beschreibung

The book offers cutting-edge insights into AI-driven optimization algorithms and their crucial role in enhancing real-time applications within fog and Edge IoT networks and addresses current challenges and future opportunities in this rapidly evolving field.

This book focuses on artificial intelligence-induced adaptive optimization algorithms in fog and Edge IoT networks. Artificial intelligence, fog, and edge computing, together with IoT, are the next generation of paradigms offering services to people to improve existing services for real-time applications. Over the past few years, there has been rigorous growth in AI-based optimization algorithms and Edge and IoT paradigms. However, despite several applications and advancements, there are still some limitations and challenges to address including security, adaptive, complex, and heterogeneous IoT networks, protocols, intelligent offloading decisions, latency, energy consumption, service allocation, and network lifetime.

This volume aims to encourage industry professionals to initiate a set of architectural strategies to solve open research computation challenges. The authors achieve this by defining and exploring emerging trends in advanced optimization algorithms, AI techniques, and fog and Edge technologies for IoT applications. Solutions are also proposed to reduce the latency of real-time applications and improve other quality of service parameters using adaptive optimization algorithms in fog and Edge paradigms.

The book provides information on the full potential of IoT-based intelligent computing paradigms for the development of suitable conceptual and technological solutions using adaptive optimization techniques when faced with challenges. Additionally, it presents in-depth discussions in emerging interdisciplinary themes and applications reflecting the advancements in optimization algorithms and their usage in computing paradigms.

Audience

Researchers, industrial engineers, and graduate/post-graduate students in software engineering, computer science, electronic and electrical engineering, data analysts, and security professionals working in the fields of intelligent computing paradigms and similar areas.

Leseprobe

Table of Contents

Cover

Table of Contents

Series Page

Title Page

Copyright Page

Preface

Acknowledgement

1 Navigating Next-Generation Network Architecture: Unleashing the Power of SDN, NFV, NS, and AI Convergence

1.1 Introduction

1.2 Revolutionizing Infrastructure with SDN, NFV, and NS

1.3 Realizing NS Potential with SDN and NFV

1.4 Artificial Intelligence: Pivotal Role in Networking Transformation

1.5 Navigating Challenges and Solutions

1.6 Conclusion

References

2 OctoEdge: An Octopus-Inspired Adaptive Edge Computing Architecture

2.1 Introduction

2.2 Problem Statement

2.3 Motivations

2.4 Related Work

2.5 OctoEdge Proposed Architecture

2.6 OctoEdge Architecture Functional Components

2.7 Results and Discussion

2.8 OctoEdge Architecture: Scope and Scientific Merits

2.9 Use Cases and Applications

2.10 Challenges and Future Directions

2.11 Conclusion

References

3 Development of Optimized Machine Learning Oriented Models

3.1 Introduction

3.2 Literature Review

3.3 Problem Definition

3.4 Proposed Work

3.5 Experimental Analysis

3.6 Conclusion

3.7 Future Scope

References

4 Leveraging Multimodal Data and Deep Learning for Enhanced Stock Market Prediction

4.1 Introduction

4.2 Literature Review

4.3 Proposed Design of an Efficient Model that Leverages Multimodal Data and Deep Learning for Enhanced Stock Market Prediction

4.4 Statistical Analysis and Comparison

4.5 Acknowledging Limitations and Potential Challenges

4.6 Mitigation Strategies and Future Directions

4.7 Conclusion

4.8 Future Scope

References

5 Context Dependent Sentiments Analysis Using Machine Learning

5.1 Introduction

5.2 Literature Review

5.3 Methodology

5.4 Proposed Model

5.5 Implementations and Results

5.6 Conclusion

References

6 Thyroid Cancer Prediction Using Optimizations

6.1 Introduction

6.2 Background and Related Work

6.3 Proposed Methodology

6.4 Architecture

6.5 Materials and Methods

6.6 Results and Discussion

6.7 Conclusion

References

7 An LSTM-Oriented Approach for Next Word Prediction Using Deep Learning

7.1 Introduction

7.2 Related Work

7.3 Design and Implementation

7.4 Proposed Model Architecture

7.5 Results and Discussions

7.6 Conclusion

References

8 Churn Prediction in Social Networks Using Modified BiLSTM-CNN Model

8.1 Introduction

8.2 Customer Behavior in Social Networks

8.3 Proposed Methodology

8.4 Result

8.5 Conclusion

References

9 Fog Computing Security Concerns in Healthcare Using IoT and Blockchain

9.1 Introduction

9.2 Related Work

9.3 Open Questions and Research Challenges

9.4 Problem Definition

9.5 Objectives

9.6 Research Methodology

9.7 Conclusion and Future Work

References

10 Smart Agriculture Revolution: Cloud and IoT-Based Solutions for Sustainable Crop Management and Precision Farming

10.1 Introduction

10.2 Data Analytics and Decision Support

10.3 Challenges and Solutions Smart Agriculture

10.4 AI for Soybean (

Glycine max

) Crop

10.5 Result Discussion

10.6 Conclusion

References

11 Greedy Particle Swarm Optimization Approach Using Leaky ReLU Function for Minimum Spanning Tree Problem

11.1 Introduction

11.2 Background

11.3 Population-Based Proposed Optimization Approach

11.4 Experimental Setup and Result Analysis of Proposed Work (LR-GPSO)

11.5 Conclusion and Future Work

References

12 SDN Deployed Secure Application Design Framework for IoT Using Game Theory

12.1 Introduction

12.2 Background Study

12.3 SDN-Deployed Design Framework for IoT Using Game-Theoretic Solutions

12.4 Case Study: SDN Deployed Design Framework in Robot Manufacturing Industry

12.5 Discussion

12.6 Conclusion

References

13 Framework for PLM in Industry 4.0 Based on Industrial Blockchain

13.1 Introduction

13.2 Related Work

13.3 The Recommended Architecture’s Methodology

13.4 Key Services That are Suggested

13.5 Modelling and Assessment

13.6 Conclusion and Future Work

A Statement of Competing Interests

References

14 Machine Learning Enabled Smart Agriculture Classification Technique for Edge Devices Using Remote Sensing Platform

List of Abbreviations

14.1 Introduction

14.2 Related Works

14.3 Methods and Dataset

14.4 Proposed Algorithm

14.5 Results and Discussions

14.6 Conclusion

References

15 A Lightweight Intelligent Detection Approach for Interest Flooding Attack

15.1 Introduction

15.2 NDN Background

15.3 Related Work

15.4 IFA Feature Selection and Detection

15.5 Conclusion

References

16 An Internet of Vehicles Model Architecture with Seven Layers

16.1 Introduction

16.2 Literature Review

16.3 Proposed Architecture of Internet of Vehicles

16.4 Applications, Characteristics, and Challenges of the Internet of Vehicles (IoV)

Conclusion

References

Index

End User License Agreement

List of Tables

Chapter 1

Table 1.1 Common SDN protocols and APIs within SDN architecture.

Table 1.2 Classification and KPIs for three use-case categories.

Table 1.3 Supervised learning algorithm for achieving the NexGen objective.

Table 1.4 Unsupervised learning algorithm for achieving the NexGen objective.

Table 1.5 Reinforcement learning algorithm for achieving the NexGen objective.

Table 1.6 A Deep learning algorithm for achieving the NexGen objective.

Table 1.7 Challenges and its solutions.

Chapter 2

Table 2.1 Comparative analysis of edge computing architectures.

Table 2.2 Comparative analysis of architecture based on QoS parameters.

Table 2.3 Scope and scientific merits of OctoEdge architecture.

Table 2.4 Scientific merits with explanation of OctoEdge architecture.

Chapter 3

Table 3.1 Attack class.

Table 3.2 Number of samples [11].

Table 3.3 Confusion matrix [12].

Table 3.4 Performance measure of models [11].

Table 3.5 Class wise performance measure [12].

Chapter 4

Table 4.1 State-of-the-art review of existing methods.

Table 4.2 Accuracy, precision, and recall comparison.

Table 4.3 F1-Score and MSE.

Table 4.4 Response time and computational efficiency.

Chapter 5

Table 5.1 Pros and Cons of the study.

Table 5.2 Text accuracy confusion matrix.

Table 5.3 Audio accuracy confusion matrix.

Chapter 6

Table 6.1 Comparison of state of art technology.

Table 6.2 Representation of evaluation matrices of proposed model.

Table 6.3 Classification techniques used and their respective factors.

Table 6.4 Experimental result of thyroid dataset on different optimizers.

Table 6.5 Registered different types of accuracy metric over compared optimiza...

Chapter 8

Table 8.1 An overview of the categories of customer behavior.

Table 8.2 Summary of the Online purchasing record.

Table 8.3 A summary of the machine learning algorithms to predict customer chu...

Table 8.4 A summary of the Deep learning algorithms to predict customer churn ...

Table 8.5 Efficiency of the proposed models.

Table 8.6 Performance analysis of modified BiLSTM-CNN models.

Table 8.7 Comparison of Machine learning models used in Churn Prediction.

Chapter 9

Table 9.1 Various threats and attacks in the domain of healthcare in fog compu...

Table 9.2 Security solutions based on fog computing in healthcare using blockc...

Chapter 10

Table 10.1 State-of-the-art comparison table of IoT and cloud in smart agricul...

Table 10.2 Challenges and solutions smart agriculture.

Table 10.3 Average running time of segmentation algorithms.

Chapter 11

Table 11.1 State of the art.

Table 11.2 Initialization of parameter for simulation.

Table 11.3 Mean value, minimum value and standard deviation of algorithms on v...

Chapter 12

Table 12.1 Signaling game model notations.

Table 12.2 Values used in experimentation.

Table 12.3 Controller action (partial) on the Raspberry Pi Screen.

Chapter 13

Table 13.1 Information about on chain/off chain using standards.

Table 13.2 Statistics of transaction speed in blockchain [51].

Table 13.3 Instrument for developing a blockchain-based PLM platform.

Table 13.4 The common comparisons in a qualitative approach with the current P...

Table 13.5 Detailed data regarding latency execution [59].

Chapter 14

Table 14.1 Sentinel 2 MSI’s confusion matrix for the MD technique.

Table 14.2 Sentinel 2 MSI’s confusion matrix for the NB technique.

Table 14.3 Overall accuracy and Kappa coefficient of Sentinel 2 MSI.

Table 14.4 CA, PA accuracy and F1 Score of Sentinel 2 MSI dataset.

Chapter 15

Table 15.1 Simulation parameters for IFA on DFN topology.

Table 15.2 Ranking of features in ascending order based of various feature lea...

Table 15.3 Average ranking of features based on Table 15.2.

Table 15.4 Feature subset selected after applying wrapper methods.

Table 15.5 Ranking of features based on Table 15.4.

Table 15.6 Classification result on full dataset.

Table 15.7 Classification result on reduced dataset.

Table 15.8 Percentage reduction in classification metrics after feature reduct...

Chapter 16

Table 16.1 Current state of the art in IoV review and research needs.

List of Illustrations

Chapter 1

Figure 1.1 (a) Decentralized traditional architecture. (b) Centralized SDN arc...

Figure 1.2 A typical architecture of SDN consists of three layers.

Figure 1.3 NFV layered architecture.

Figure 1.4 NS framework.

Figure 1.5 Illustration of integration of SDN and NFV.

Chapter 2

Figure 2.1 OctoEdge architecture flow diagrams.

Chapter 3

Figure 3.1 Proposed model.

Figure 3.2 Protocol type frequency.

Figure 3.3 Relationships between protocol type, assault, and duration.

Figure 3.4 Statistics of SVM model with confusion matrix.

Figure 3.5 Confusion matrix and statistics of XGBoost model.

Figure 3.6 Performance measure [7].

Figure 3.7 Class wise comparison.

Chapter 4

Figure 4.1 Overall flow of the proposed model for multimodal stock predictions...

Figure 4.2 Pseudo code for the proposed model for stock predictions.

Chapter 5

Figure 5.1 System architecture.

Figure 5.2 Text cleaning pipeline.

Figure 5.3 Audio cleaning and conversion.

Figure 5.4 Home page.

Figure 5.5 Audio sentiment home page.

Figure 5.6 Audio sentiment accuracy curve.

Figure 5.7 Audio sentiment loss curve.

Figure 5.8 Emotion detected (neutral).

Figure 5.9 Line chart for varying emotions.

Figure 5.10 Line chart for varying emotions.

Figure 5.11 Xception accuracy graph.

Figure 5.12 Xception loss graph.

Chapter 6

Figure 6.1 The histology of thyroid malignancies.

Figure 6.2 An outline of the research techniques used in this proposal.

Figure 6.3 Distribution of various factors.

Figure 6.4 CNN training procedure for hypo thyroid and non-hypo thyroid.

Figure 6.5 Flowchart of the prognosis stage in the proposed approach of machin...

Figure 6.6 A pictorial representation of the performance of the proposed model...

Figure 6.7 A pictorial representation of the various types of accuracy obtaine...

Figure 6.8 (a) Adam optimizer.

Figure 6.8 (b) Adadelta optimizer.

Figure 6.8 (c) SGD optimizer.

Figure 6.8 (d) Adagrad optimizer.

Figure 6.8 (e) RMSprop optimizer.

Chapter 7

Figure 7.2 Architecture of Bi-LSTM model.

Figure 7.3 Architecture of workflow of next word prediction.

Figure 7.4 Model parameter of LSTM.

Figure 7.5 Model parameters of Bi-LSTM.

Figure 7.6 Training and testing loss of data for LSTM model.

Figure 7.7 Training and testing accuracy of data for LSTM model.

Figure 7.8 Training accuracy of data for Bi-LSTM.

Figure 7.9 Training loss of data for Bi-LSTM.

Figure 7.10 Prediction result.

Chapter 8

Figure 8.1 Predictive analytics areas.

Figure 8.2 Applications in social network analysis.

Figure 8.3 A framework of customer behavior in predictive analytics.

Figure 8.4 Flowchart of proposed algorithm for customer churn prediction.

Figure 8.5 Efficiency of the proposed models.

Figure 8.6 Performance analysis of modified BiLSTM-CNN models.

Figure 8.7 Comparison with machine learning models.

Figure 8.8 The recommended framework compared to deep learning models.

Chapter 9

Figure 9.1 Fog-enabled IoT systems.

Figure 9.2 Functionality-based fog architecture.

Figure 9.3 The three-layer design of blockchain integration with IoT.

Figure 9.4 The passage gathers information from a gadget.

Figure 9.5 Gateway adds a new square to the record of B.C. in layer two.

Figure 9.6 Proposed framework.

Chapter 10

Figure 10.1 Smart agriculture [34].

Figure 10.2 Clouds computing in agriculture [35].

Figure 10.3 Precision farming [36].

Figure 10.4 The role of AI in the agriculture information management [37].

Figure 10.5 Plants disease a pets detection using AI technology [38].

Figure 10.6 Conceptual diagram and working model of crop health monitoring [39...

Figure 10.7 The removing-background images with grabcut algorithm [40].

Figure 10.8 Flowchart of disease spot image segmentation algorithm.

Figure 10.9 Segmentation effect of different algorithms on the brown spot dise...

Chapter 11

Figure 11.1 Basic structure of LRGPSO.

Figure 11.2 Flowchart of LRGPSO.

Figure 11.3 Results of simulation on vertices (V=20).

Figure 11.4 Results of simulation on vertices (V=40).

Figure 11.5 Results of simulation on vertices (V=60).

Figure 11.6 Results of simulation on vertices (V=80).

Figure 11.7 Convergence curve.

Chapter 12

Figure 12.1 Simplified IoT architecture consists of three different layers.

Figure 12.2 The three levels that make up SDN architecture are reachable via o...

Figure 12.3 Proposed SDN-deployed design framework for IoT.

Figure 12.4 The controller’s representation of the trust computation. An intru...

Figure 12.5 Strategy of sender/sensor: behaves

Malicious

if it demands more in...

Figure 12.6 Packets found using both the signaling game model and not.

Figure 12.7 Calculated belief value of the malicious and legitimate sensor.

Chapter 13

Figure 13.1 The architecture of the proposed blockchain-based PLM [52].

Figure 13.2 The data transmission from a blockchain network to a machine level...

Figure 13.3 Stages of product life cycle [54].

Figure 13.4 The co-creation blockchain-based collaborative development.

Figure 13.5 Real-time tracking and tracing service enabled by blockchain techn...

Figure 13.6 Preventative maintenance service powered by blockchain [56].

Figure 13.7 Recycling mechanism for blockchain-based design.

Figure 13.8 The smart contract code sample for rist analytics [57].

Figure 13.9 The differing block sizes’ latency under varying transaction arriv...

Figure 13.10 ET and the suggested platform: a comparative analysis.

Chapter 14

Figure 14.1 Area of interest (study area map).

Figure 14.2 Proposed methodology.

Figure 14.3 Classified map of NB and MD classifiers.

Chapter 15

Figure 15.1 Interest and data packet.

Figure 15.2 NDN forwarding pipeline.

Figure 15.3 DFN topology.

Figure 15.4 Snapshot of the dataset.

Chapter 16

Figure 16.1 Internet of Vehicles interaction model.

Figure 16.2 Internet of Vehicles network model.

Figure 16.3 Internet of Vehicles environmental model.

Figure 16.4 Architecture of Internet of Vehicles.

Guide

Cover Page

Table of Contents

Series Page

Title Page

Copyright Page

Preface

Acknowledgement

Begin Reading

Index

WILEY END USER LICENSE AGREEMENT

Pages

ii

iii

iv

xv

xvi

xvii

xviii

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

61

62

63

64

65

66

67

68

69

70

71

72

73

74

75

76

77

78

79

80

81

82

83

84

85

86

87

88

89

90

91

92

93

94

95

96

97

98

99

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

224

225

226

227

228

229

230

231

232

233

234

235

236

237

238

239

240

241

242

243

244

245

246

247

248

249

250

251

252

253

254

255

256

257

258

259

260

261

262

263

264

265

266

267

268

269

270

271

272

273

274

275

276

277

278

279

280

281

282

283

284

285

286

287

288

289

290

291

292

293

294

295

296

297

298

299

300

301

302

303

304

305

306

307

308

309

310

311

312

313

314

315

316

317

318

319

320

321

322

323

324

325

326

327

328

329

330

331

332

333

334

335

336

337

338

339

340

341

342

343

344

345

346

347

348

349

350

351

352

353

354

355

356

357

358

359

360

361

362

363

364

365

366

367

368

369

370

371

372

373

374

375

376

377

378

379

380

381

382

383

384

385

386

387

388

389

390

391

392

393

394

395

396

397

398

399

400

401

402

403

404

405

406

407

408

409

410

411

412

413

414

415

416

417

418

419

420

421

422

423

424

425

426

427

428

429

430

431

432

433

434

435

436

437

438

439

440

441

442

443

444

445

446

447

448

449

450

451

452

453

454

455

456

457

458

459

460

461

Scrivener Publishing100 Cummings Center, Suite 541JBeverly, MA 01915-6106

Machine Learning in Biomedical Science and Healthcare Informatics

Series Editors: Vishal Jain ([email protected])and Jyotir Moy Chatterjee ([email protected])

In this series, an attempt has been made to capture the scope of various applications of machine learning in the biomedical engineering and healthcare fields, with a special emphasis on the most representative machine learning techniques, namely deep learning-based approaches. Machine learning tasks are typically classified into two broad categories depending on whether there is a learning ‘label’ or ‘feedback’ available to a learning system: supervised learning and unsupervised learning. This series also introduces various types of machine learning tasks in the biomedical engineering field from classification (supervised learning) to clustering (unsupervised learning). The objective of the series is to compile all aspects of biomedical science and healthcare informatics, from fundamental principles to current advanced concepts.

Publishers at ScrivenerMartin Scrivener ([email protected])Phillip Carmical ([email protected])

AI-Based Advanced Optimization Techniques for Edge Computing

Edited by

Mohit Kumar

Dept. of Information Technology, Dr. B R Ambedkar National Institute of Technology, Jalandhar, India

Gautam Srivastava

Dept. of Mathematics & Computer Science, Brandon University, Manitoba, Canada

Ashutosh Kumar Singh

Dept. of Computer Science and Engineering, United College of Engineering & Research, Allahabad, India

and

Kalka Dubey

Dept. of Computer Science and Engineering, Rajiv Gandhi Institute of Petroleum Technology, Amethi, India

This edition first published 2025 by John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030, USA and Scrivener Publishing LLC, 100 Cummings Center, Suite 541J, Beverly, MA 01915, USA© 2025 Scrivener Publishing LLCFor more information about Scrivener publications please visit www.scrivenerpublishing.com.

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, except as permitted by law. Advice on how to obtain permission to reuse material from this title is available at http://www.wiley.com/go/permissions.

Wiley Global Headquarters111 River Street, Hoboken, NJ 07030, USA

For details of our global editorial offices, customer services, and more information about Wiley products visit us at www.wiley.com.

Limit of Liability/Disclaimer of WarrantyWhile the publisher and authors have used their best efforts in preparing this work, they make no representations or warranties with respect to the accuracy or completeness of the contents of this work and specifically disclaim all warranties, including without limitation any implied warranties of merchantability or fitness for a particular purpose. No warranty may be created or extended by sales representatives, written sales materials, or promotional statements for this work. The fact that an organization, website, or product is referred to in this work as a citation and/or potential source of further information does not mean that the publisher and authors endorse the information or services the organization, website, or product may provide or recommendations it may make. This work is sold with the understanding that the publisher is not engaged in rendering professional services. The advice and strategies contained herein may not be suitable for your situation. You should consult with a specialist where appropriate. Neither the publisher nor authors shall be liable for any loss of profit or any other commercial damages, including but not limited to special, incidental, consequential, or other damages. Further, readers should be aware that websites listed in this work may have changed or disappeared between when this work was written and when it is read.

Library of Congress Cataloging-in-Publication Data

ISBN 978-1-394-28703-1

Front cover images courtesy of Adobe Firefly.Cover design by Russell Richardson

Preface

This book was written to bridge the gap between existing state-of-the-art technologies and the evolving requirements of modern industries. It provides emerging research that explores both theoretical and practical aspects of implementing new and innovative intelligent techniques across a variety of sectors, including Edge Computing, Cloud Computing, the Internet of Things, Agriculture, and Artificial Intelligence. This book serves as a valuable resource for academics, IT specialists, industry professionals, researchers, engineers, and authors seeking insights into emerging trends in AI-enabled Cloud and Edge Computing for IoT applications. It aims to explore the intricate relationship between AI and Edge/Cloud computing, delving into their synergies, applications, and future implications.

This book comprises 16 chapters, each covering intertwining concepts at two key levels of interest to the scientific community: Artificial Intelligence and Edge/Cloud Computing.

Chapter One explores navigating next-generation network architecture, unleashing the power of SDN, NFV, NS, and AI convergence. Chapter Two examines OctoEdge, an octopus-inspired adaptive edge computing architecture. Chapter Three discusses the development of optimized machine learning-oriented models.

Chapter Four focuses on leveraging multimodal data and deep learning for enhanced stock market prediction. Chapter Five delves into context-dependent sentiment analysis using machine learning. Chapter Six investigates enhancing thyroid cancer prediction by applying machine learning algorithms to clinical data.

Chapter Seven presents an LSTM-oriented approach for next-word prediction using deep learning. Chapter Eight analyzes churn prediction in social networks using a modified BiLSTM-CNN model. Chapter Nine addresses security concerns in healthcare fog computing using IoT and blockchain.

Chapter Ten highlights the smart agriculture revolution with cloud and IoT-based solutions for sustainable crop management and precision farming. Chapter Eleven explores a greedy particle swarm optimization approach using the Lecky ReLU function for solving minimum spanning tree problems.

Chapter Twelve introduces an SDN-deployed secure application design framework for IoT using game theory. Chapter Thirteen presents a framework for PLM in Industry 4.0 based on industrial blockchain.

Chapter Fourteen discusses a machine learning-enabled smart agriculture classification technique for edge devices using a remote sensing platform. Chapter Fifteen examines a lightweight intelligent detection approach for interest flooding attacks. Chapter Sixteen describes an Internet of Vehicles model architecture with seven layers.

This book may serve as a reference for a graduate course in Artificial Intelligence and Cloud Computing. Readers are expected to be well-versed in the basic concepts of Machine Learning, Distributed Computing, and the Internet of Things. The theoretical concepts presented will be valuable for coursework.

Writing this book has been a rewarding experience, made possible by the tremendous efforts of a dedicated team. We extend our gratitude to the authors who contributed their respective chapters, as well as to the editors who offered valuable suggestions for improving content delivery. Every piece of feedback was carefully considered, and it has undoubtedly shaped parts of the work. We are especially grateful to Martin Scrivener and Scrivener Publishing for their help and publication. Finally, we thank our families for their unwavering support—without them, this book would not have been possible.

November 2024

Acknowledgement

The writing of this book has been a rewarding experience and elaborates a huge effort from a team of very dedicated contributors. We would like to thank list of authors who contributes their respective chapter and we are also thankful to the list of editors who provides suggestions for better delivery of content. All feedback was considered and there is no doubt that there will be some content influenced by the suggestions. We especially thank to the publisher who believes in the content and provides a platform to reach it out to the audience. Finally, we are thankful to our family for their continued support. Without them, the book would not have been possible.

1Navigating Next-Generation Network Architecture: Unleashing the Power of SDN, NFV, NS, and AI Convergence

Monika Dubey1*, Snehlata2, Ashutosh Kumar Singh2, Richa Mishra1 and Mohit Kumar3

1Department of Electronics & Communication, University of Allahabad, Prayagraj, U.P., India

2Department of Computer Science and Engineering, United College of Engineering & Research, Prayagraj, U.P., India

3Department of Information Technology, National Institute of Technology Jalandhar, Punjab, India

Abstract

The framework for existing legacy network architecture is massive and complex. It mainly relies on inflexible and expensive equipment, typically constructed from a massive number of switches, routers, firewalls, and hubs. Moreover, this vendor-specific network configuration and complex control protocols are not flexible enough to offer customized quality of services (QoS). Provisioning of next-gen (Next Generation, 5G, and beyond) technologies, software-defined networking (SDN), network function virtualization (NFV), and network slicing (NS) work as catalysts to offer simplified, customized, and clever networking. To provide centralized positioning, SDN decouples the control plane (CP) and data plane (DP) from the traditional router. In the SDN architecture, decision making and network control are now done at a centralized place known as the controller. However, DP is still intact with the routing device. This arrangement privileges the network administrators to control, manage, and alter network behavior dynamically. To contrast the vender-specific networking, NFV allows network functions (NFs) to run on generic hardware. In this direction, NS pioneers QoS-specific use cases as a new business model. NS involves the slicing of a single physical network in the form of multiple slices. It not only supports the customization of QoS services for diverse use cases, but it also improves isolation, independence, multitenancy, dynamic resource allocation, and end-to-end service provisioning. In this chapter, we first delved into NexGen’s promising technologies and explored their intertwined role and impact on the modern networking framework. We accessed various SDN and NFV architectures and discussed network-slicing framework. Secondly, we have shed light on the importance of AI-driven automated network management over traditional network approaches. In this sequence, we conducted a comparative analysis of AI-driven machine learning (ML) and deep learning (DL) approaches in the context of NextGen technologies. In this chapter, we intend to systematically and intricately navigate the multifaceted landscape of NexGen technologies. This chapter will offer researchers, industry stakeholders, and practitioners a timely and deeper understanding of transformative technology and its impact on modern network paradigms.

Keywords: Next-generation technology, SDN, NFV, QoS, NS

1.1 Introduction

The evolution of network technologies has marked pivotal advancements in the telecom sector. It spans from the radiant stage of ARPANET to modern networking. The existing legacy network architecture is based upon un-flexible and costly network equipment comprising switches, hubs, routers, and firewalls [1]. These proprietary hardware-based traditional networks grapple with the demands of modern networking. The surge of extensive data traffic, dynamic network conditions, and the need for real-time decision-makers pose challenges that traditional networks are not capable of addressing efficiently [2]. Traditional methods, such as Static Routing, Ethernet, Transmission Control Protocol (TCP), and Internet Protocol (IP), are built on manual configuration and static protocols. With the surge of diverse applications, customized QoS, high volume, and unpredicted traffic necessitate a paradigm shift. To address these limitations of the traditional approach, Next-Gen (Next Generation, 5G, and beyond) technologies, Software Defined Networking (SDN), Network Function Virtualization (NFV), and NS act as catalysts for redefining the network paradigm. SDN [3] disrupts traditional decentralized architecture by decoupling the Control Plane (CP) and Data Plane (DP) from conventional routers. This centralized control and decision-making entity is known as the controller. This architectural shift empowers the network controller to dynamically manage, control, and modify the network behavior. Concurrently, NFV [4] revolutionizes network functionality by enabling them to run on generic hardware instead of proprietary hardware, offering cost-effectiveness, flexibility, and simplified maintenance. With the advancement of the network landscape, customize QoS-specific servers are the new business model. In this direction, NS [5] has become a revolutionary approach, involving the partitioning of a single physical network into multiple slices. It not only offers customized QoS requirements to modern applications but also enhances isolation, dynamic resource allocation, multi-tenancy, and security [6].

This book chapter also explored the NextGen promising technologies and their intertwined role and impact on modern networking. Traditional networking approaches are static and require human intervention during changes in the network. The increase in network size and the unpredictable nature of network traffic make them more time-consuming and complex. Therefore, AI emerges as a key driver for NextGen networking. It introduced the level of intelligence with its learning and capability of predictive analysis. This chapter also sheds light on how AI-driven approaches complement and enhance the functionalities of SDN, NFV, and NS.

The contributions and highlight of this book chapter are as follows:

Initially, we present a concise overview of the evolutionary history of network technologies and the key phases that shaped the modern networking landscape.

To explore the transformative NexGen technologies (SDN, NFV, and NS), we highlight the influence and intertwining role of NexGen technologies.

This paper systematically highlights the importance of AI over traditional methods. In this sequence, we conducted a comparative analysis of AI-driven Machine Learning (ML) and Deep Learning (DL) approaches in the context of NextGen technologies.

Finally, we identify challenges associated with NexGen Technologies and with the integration of these modern technologies.

In a nutshell, this chapter will offer researchers and industry stakeholders a timely and deep understanding of transformative NexGen technologies and the impact of their combination on modern technology. It also includes the contribution and comparative analysis of AI-driven algorithms in the context of NexGen technologies.

1.2 Revolutionizing Infrastructure with SDN, NFV, and NS

Due to increasing day-to-day network traffic, networking technologies have undergone a continuous evolution, and based on this, they can be categorized into several phases, such as traditional networking, Wireless Sensor Networking (WSN), client-server networking, and more. Before discussing NexGen technologies and its specifications, it is crucial to examine the evolutionary changes of networking technologies and the key developments that have been influenced by traditional networking. Concise overview is given as follows:

ARPANET and Early Networking:

ARPANET:

The Advanced Research Projects Agency Network (ARPANET) [

7

], established in the 1960s, conducted early experiments for linking computer systems over short distances. It laid the foundation for modern networking. However, these networks remained restricted to research institutions.

Packet Switching:

The development of packet switching [

8

], a key innovation during this era, allowed data to be broken into packets, transmitted independently, and reassembled at the intended destination.

The pioneering work and packet switching laid the fundamental groundwork for the internet.

Emergence of the Internet:

Standardization (TCP/IP):

During the 1980s, the TCP [

9

] and IP underwent standardization, forming the backbone of the modern Internet.

Commercialization:

The Internet underwent a pivotal shift from being primarily dedicated to research and academia to a commercial platform, leading to the rise of the World Wide Web (WWW). It establishes the fundamental framework for the contemporary Internet.

Emergence of Client-Server Architecture and LANs:

Client-Server Model:

In 1980s, the paradigm of computing is shifting from centralized mainframes to distributed systems with the client-server model [

10

].

The rise of Local Area Networks (LANs):

The internet and other LAN technologies emerged, allowing computers to share resources within confined spaces.

Wireless Networking and Mobility:

Wi-Fi Standardization:

In the 2000s, the standardization of wireless technologies, particularly Wi-Fi adoption [

11

], empowered enhanced mobility and flexibility in network access.

Expansion of Mobile Networks:

The surge in mobile device usage during this era empowered enhanced mobility and flexibility in network access [

12

].

Cloud Computing and Virtualization:

Evolution of Cloud Services:

The 2010s witnessed a transformative shift with the advent of cloud computing [

13

], fundamentally changing the way data and applications are stored and accessed.

Rise of Virtualization:

The decade also saw the emergence of NFV and SDN [

14

], contributing to enhanced flexibility and efficiency in the management of network resources.

In the beginning, traditional enterprise networks followed conventional decentralized designs and scattered collections of purpose-built routers, switches, and middle-boxes supplied by various hardware vendors [15]. Each device uses embedded proprietary hardware and logic to make forwarding decisions, filter traffic, or transform flows. This distributed CP closely relates key networking functions to the restrictions of the underlying boxes in terms of capability and flexibility. The conventional decentralized networking architecture imposed significant barriers to change in network arrangements. Every configuration change or new policy meant navigating vendor-specific command-line interfaces to manually reprogram individual pieces of equipment. To deal with the huge dynamic traffic, this fragmented model is not appropriate due to the rigidities of closed hardware systems. The massive burden of managing numerous devices running complex embedded protocols eventually became unsustainable. To address the longstanding limitations of traditional network architectures, the SDN paradigm emerged to unlock network flexibility and fundamentally introduce centralized network control [16]. The architectural differences between traditional decentralized architecture and centralized architecture are presented in Figure 1.1(a) and 1(b) respectively.

Figure 1.1 (a) Decentralized traditional architecture. (b) Centralized SDN architecture.

1.2.1 SDN: Definition and Architecture

SDN architecture [17] is the paramount approach for centralized network control. It is structured to decouple the control plane from the data plane and provides automation and centralized control by delegating specialized functions to each level via programmatic APIs. The centralized control unit, known as the controller, is responsible for network design, decision-making, and network management. The SDN architecture typically comprises three main components:

Application Layer:

The topmost layer of SDN consists of software programs that communicate business-related policy and network behavior. This layer interacts with the SDN controller to communicate policies, requirements, or network changes. Common SDN applications include load balancing, traffic monitoring, and security applications.

Control Layer:

This intermediary layer, known as the SDN controller, is the brain of the SDN architecture. The controller communicates with network devices in the infrastructure layer via southbound and SDN applications via northbound APIs at the application layer.

Infrastructure Layer:

The bottom layer is the infrastructure layer, which consists of the physical and virtual network devices those forward data packets. In contrast to traditional networking by separating the CP, the intelligence for decision-making is moved from individual devices to the centralized controller.

The architecture presented in Figure 1.2 outlines the fundamentals of SDN. However, SDN provides incredible versatility to adapt its core principles into diverse architectural designs to address specific networking needs and challenges. In the realm of single-layer architectures, centralized controller manages the entire network, whereas distributed SDN architecture’s [18] CP functions across multiple controllers to provide more scalability in comparison with a single SDN controller. Multi-layer SDN architecture presents hierarchical SDN architecture [19] with multiple layers of controllers. It helps to enhance organization and management in large-scale networks. On the other hand, in a hybrid SDN architecture, SDN coexists with traditional networking (NON-SDN) [20] elements. It allows for seamless integration of SDN principles with traditional networking elements, allowing coexistence and transition. Overlay SDN architectures [21] are commonly prevalent in data centre environments. In this tunneling, protocols are used to create virtual networks on top of the physical infrastructure. Cloud SDN architectures [22] focus on cloud environments, emphasizing automation, agility, and the ability to adapt to the dynamic workloads characteristic of cloud computing. Intent-Based Networking (IBN) architectures [23] are mainly focused on high-level business intent for automated and simplified management of networks on the basis of desired output. Tailored for 5G networks, the 5G SDN architecture integrates SDN with NFV to meet the demands of next-generation network framework. SDN protocols play a crucial role in communication and coordination between various components of SDN. It primarily facilitates communication between components, policy dissemination, dynamic adoption, load balancing, and configuration management. Table 1.1 outlines the common SDN protocols and APIs within SDN architecture.

Figure 1.2 A typical architecture of SDN consists of three layers.

Table 1.1 Common SDN protocols and APIs within SDN architecture.

Aspect

SDN protocol/API

Description

Northbound APIs

Open Flow

A standard protocol between SDN controllers and network devices such as switches to define flows.

REST APIs

Representational State Transfer (REST) APIs leverage controller communication with SDN applications for northbound interactions.

NETCONF

It is used for northbound communication between controllers and network devices for network management.

Southbound APIs

Open Flow

Southbound protocol, communicate between SDN controllers and network switches to configure data plane behavior.

P4 (Programming Protocol-Independent Packet Processors)

Southbound API for defining packet forwarding behaviors to define packet processing across devices.

East-West APIs

VXLAN

Facilitating east-west traffic to create virtual overlay networks across data centers.

Geneve

Another overlay protocol for east-west communication for network virtualization across SDN environments.

1.2.2 NFV: Definition and Architecture

Non-virtualized traditional networks run on dedicated proprietary hardware. Unlike them, NFV supports the sharing of infrastructure resources during NF deployments and runs as a software application on generic hardware instead of proprietary hardware. It virtualizes NFs such as firewalls, routers, and load balancers, also known as VNFs (Virtual NFs). The NFV architectural framework defined by ETSI [24] consists of three key domains:

Virtualized Network Functions (VNFs):

VNFs are software applications implemented on network functions to replace dedicated appliances. These software instances replicate the functionality of traditional network devices such as firewalls and load balancers.

NFV Infrastructure (NFVI):

This includes the infrastructure components (compute, storage, and networking; Commercial-off-the-Shelf (COTS) hardware like servers, switches, and storage deployed in data centers); and the virtual layer on which VNFs run.

NFV Management and Orchestration (NFV MANO):

This includes orchestrators, VNF managers, and it supports the framework for orchestration and management of the lifecycle of VNFs across the NFVI.

ETSI defines the foundational NFV architectural block presented in Figure 1.3. However, the NFV architecture exhibits diverse forms to accommodate diverse scenarios and specific operational requirements. In a centralized NFV architecture [19], management and orchestration functions are consolidated to simplify CP. However, centralized designs focused exclusively on operational efficiency can suffer from latency limitations in distributed deployments. Meanwhile, distributed NFV infrastructure [25] spreads capabilities across multiple localized data centers, catering to scenarios where low-latency communication is critical, as seen in edge computing environments. Hybrid architecture is intended to balance the tradeoffs between centralized and distributed architecture. In this architecture, common network functions get consolidated into a core virtualized infrastructure for efficiency, while other specialized functions continue at the edge for performance.

Figure 1.3 NFV layered architecture.

1.2.3 NS: Conceptual Abstractions

Traditional mobile networks are based on the “one-size-fits-all” network paradigm [26]. It is no longer efficient to support different use cases. Third-generation partnership project (3GPP) [27] defined NS is a key concept for 5G networks to offer flexible and customized network services. In this approach, “slicing” is the concept of dividing physical network resources into logical networks in the form of slices to address different QoS and Service Level Agreements (SLAs) [28]. It involves logically partitioned physical network infrastructure to offer highly customized and optimized connectivity solutions for various use cases. End-to-End (E2E) NS [29] involves the orchestration and customization of resources across the entire network architecture. In the 5G architecture, it is logically partitioned across different components of the network Core Network (CN), Radio Access Network (RAN), and Transport Network (TN) domains [30].

Core NS:

Core NS has a dedicated virtual core network, and central control mechanisms orchestrate the flow of data and session management services. It supports distinct use case applications ranging from augmented reality to mission-critical communications in terms of key network performance such as latency, bandwidth, jitter, reliability, and so on.

Transport NS:

It provides connectivity between the core network and RAN elements of each slice for seamless movement of data between various network elements. It ensures dedicated traffic capacity and guaranteed resources like bandwidth, latency, reliability, and jitter for each network slice.

Radio Access Network (RAN) Slicing:

RAN provides the interface between the user device and the broader network. RAN slicing is responsible for fulfilling the user demands of different use cases by isolating radio resources, including bandwidth and frequencies. It is mainly crucial for scenarios where varied performance characteristics are in demand, such as high bandwidth and ultra low latency. It plays a significant role in addressing vertical industry service-level requirements in terms of network performance (latency, throughput, reliability, etc.) for diverse use-case services. In this context, more than 400 vertical use cases [

31

] were identified across various industries for 5G, such as smart cities [

32

], smart factory [

33

], health care [

34

], autonomous vehicles [

35

], and so on. ITU classified these diverse use cases in terms of three broad service categories [

36

]:

(i)

eMBB (enhanced Mobile Broadband);

(ii)

massive Machine Type Communication (mMTC); and

(iii)

ultra-Reliable and Low Latency Communication (uRLLC). ITU defined this in terms of eight Key Performance Indices (KPIs) [

37

], including performance indicators (peak data rate, user experience data rate, latency, connection density, traffic volume density, and mobility) and efficiency indicators (spectrum efficiency and energy efficiency). Classification and KPI for three service categories are represented in

Table 1.2

. The network architecture and key use case scenario presented in

Figure 1.4

.

Table 1.2 Classification and KPIs for three use-case categories.

Characteristics

eMBB

mMTC

uRLLC

Aim

Human-centric data driven use cases

Multimedia content

Service provider centric

Massive connection devices

Low cost and non-delay sensitive devices

Network operator centric

Delay sensitive services

Use Case

AR/VR, real-time gamming, hotspot and wide area coverage.

Smart wearable, smart city, smart power grid, smart industries

V2X, public safety, remote surgery, smart industries and robotics

Parameters

Peak data rate

20 GBPS

Connection density

106 devices/km

2

U-plane latency

1ms

C-plane latency

10ms

U-plane latency

4ms

Area traffic capacity

>3 times greater than LTE advanced

C-plane latency

20ms

User Equipment data rate

>3 times greater than LTE advanced

UE battery life

Beyond 10 years

Reliability

10

-5

for 32 Bytes (user plane latency of 1ms)

Target mobility speed

500 Km/h

Mobility interruption time

0ms

Figure 1.4 NS framework.

1.3 Realizing NS Potential with SDN and NFV

SDN and NFV are the building blocks of modern technology. It plays a crucial role in meeting the diverse service requirements of various use cases. Together, these technologies create powerful and flexible network architectures. These two advanced technologies introduce a new paradigm for designing, optimizing, and offering scalability, flexibility, and efficiency [38]. Key aspects where SDN and NFV complement each other are discussed below:

Centralized Network Control:

The Centralized architecture of SDN allows the network administrator to create and manage network resources dynamically through the centralized controller. NFV complements SDN by creating VNFs, allowing them to run as software on generic hardware. It enables the SDN controller to dynamically orchestrate the deployment and chaining of these VNFs.

Dynamic Resource Allocation:

SDN contributes to the dynamic allocation of resources by decoupling the CP from the DP. It allows the adjustment of network paths and traffic flow on the fly. On the other hand, NFV helps enhance resource utilization by running NFs as virtual instances.

Service Chaining:

SDN contributes to the creation of service chains in the predefined order. Whereas, NFV enhances the flexibility in service chaining. SDN controllers play a pivotal role in orchestrating these service chains.

Flexibility and Agility:

Software-defined policies allow rapid changes in network configuration. NFV brings flexibility by decoupling network functions from dedicated hardware.

End-to-End Network Orchestration:

SDN contributes to orchestrating network elements within the data center and across wide-area networks. The interplay of SDN and NFV provides end-to-end orchestration capabilities for dynamic service delivery.

In summary, SDN provides centralized abstractions of the overall network that represent in Figure 1.5. On the other hand, NFV transforms the deployment of core network functions like authentication, policies, and routing into modular software services.

NS is proven to be a revolutionary technology for unlocking highly customized and adoptable modern networks. Integration of NS, SDN, and NFV heralds a transformative era in modern telecommunication [39]. NS allows the creation of customized virtualized networks to address the unique requirements of diverse use cases. When coupled with SDN, it enables dynamic instantiation of NS and rapid customization and adoption of change. NS is intended to allocate dedicated resources to each slice by ensuring KPIs for specific user cases in terms of bandwidth, low latency, and reliability. Integration of NFV complements NS by enabling flexible deployment of VFs to ensure resource efficiency. SDN’s orchestration capabilities support NS and E2E service orchestration. Isolation in NS ensures security between slices that run on common physical infrastructure. SDN’s centralized control uses stringent security policies to enforce security and consistency across all slices. Integration of NFV complements the major objective of NS to support diverse use case requirements ranging from smart cities to automated vehicles.

Figure 1.5 Illustration of integration of SDN and NFV.

In conclusion, NS is positioned as a fundamental enabler of 5G for customized telecom services. In conclusion, NS is positioned as a fundamental enabler of 5G for customized telecom services. The combination of NS, SDN, and NFV works as the foundation for evolving network architecture. It ensures customizations, scalability, and adoption in the 5G era and beyond.

1.4 Artificial Intelligence: Pivotal Role in Networking Transformation

With the shift in the networking paradigm, traditional networking has become inefficient to deal with large, dynamic, and heterogeneous network infrastructure. The key challenges of traditional methods to address modern networking technology are as follows:

A Traditional network requires manual intervention for any network configuration or policy change.

Troubleshooting with traditional methods is often reactive and requires administrators to address this issue when it occurs.

There is a lack of predictive analysis capability to proactively analyze any issue.

Inflexible scaling is another major limitation of traditional methods. It becomes time-consuming and requires manual adjustments and upgrades to the hardware.

Modern large, dynamic, and heterogeneous networks demand quick changes in traffic patterns and user behavior. Traditional networking struggles to do so.

In summary, traditional networking approaches often involve manual and static configuration and reactive approaches. AI is proven to be a catalyst for modern network complexities [40]. AI-driven approaches introduce automation, flexibility, scalability, and intelligence. NextGen technologies such as SDN, NFV, and NS, with the integration of AI [41]; better align with the requirements of contemporary modern network environments. For SDN, AI drives capabilities, including load balancing, centralized networking control, dynamic routing, and anomaly detection and so on [42]. Regarding NFV, emphasize its ability to deal with virtualizing network functions, service chain management, resource optimization, network customization, and so on [43]. AI embedded with NS contributes to supporting customized QoS, isolation, dynamic resource allocation, multi-tenancy, and E2E service provisioning [44].

AI-embedded approaches can further be classified into ML and DL approaches; the ML approach can further be divided into supervised, unsupervised, and reinforcement learning [45]. DL is considered a specialized subset of ML. In this section, we are intended to investigate AI-driven approaches to address modern networking objectives. In particular, this section is intended to explore Ml and DL approaches to addressing SDN, NFV, and NS objectives.

1.4.1 Supervised Learning

Supervised learning is the method of making predictions or decisions based on training a model on a labeled dataset. Supervised learning can be expressed as labeled dataset D, represented in equation (1.1);

(1.1)

where xn and yn represent input feature and corresponding labels, respectively. By identifying relationships between patterns and data model maps F, represented in equation (1.2).

(1.2)

In this process, the model learns by adjusting parameters to minimize the loss function L, represented in equation (1.3).

(1.3)

After completion of the training process model, it will finally be able to predict new and unseen data.

In the context of networking, input could present various networking parameters such as performance metrics (bandwidth, latency, and so on), network traffic, or device state. While the output could denote a specific action or classification, It mainly employed for various networking applications such as traffic classification, QoS optimization and anomaly detection. Supervised learning plays a crucial role in enhancing the capabilities of SDN, NFV, and NS. Decision Tree (DT), Support Vector Machine (SVM), k-Nearest Neighbors (k-NN) and Random Forest (RF) are promising approaches to address specific objectives and challenges within the networking paradigm. The supervised learning method and its application in context of modern technologies are summarized in Table 1.3.

Table 1.3 Supervised learning algorithm for achieving the NexGen objective.

Technology

Supervised learning algorithms

Use case

References

SDN

Decision Trees

Traffic engineering, policy-based decision-making, security

[

46

,

47

]

SVM

Traffic classification, anomaly identification, intrusion detection,

[

48

,

49

]

k-Nearest Neighbors (k-NN)

network optimization, load balancing,

[

42

,

50

]

Random Forests

Dynamic routing, traffic classification, anomaly mitigation

[

42

,

51

,

52

]

Logistic Regression

network security, Policy enforcement,

[

53

,

54

]

NFV

Decision Trees

Service chain management, resource allocation

[

55

,

56

]

SVM

Fault management

[

57

]

k-Nearest Neighbors (k-NN)

Scaling. interoperability

[

58

]

Random Forests

Fault tolerance, auto-scaling VNFs

[

59

]

SVM

Network traffic classification

[

60

]

k-Nearest Neighbors (k-NN)

Resource allocation

[

61

]

1.4.2 Unsupervised Learning

Unsupervised learning is the method of training a model on an unlabeled dataset. Unsupervised learning can be expressed as: Unlabeled dataset D represented in equation (1.4);

(1.4)

Where xn represents the input patterns. The unlabeled dataset algorithm is intended to uncover hidden patterns and relationships within the data to identify hidden patterns in the dataset or group the elements of the dataset. This approach is commonly used for clustering and dimension reduction.

In the context of networking, it is manually utilized for identifying irregular traffic patterns to indicate potential security threads. It is also applicable to traffic analysis and data compression. Unsupervised algorithms play an important role in fulfilling the diverse objectives of modern networking. Clustering algorithms, such as K-mean, are applicable for identifying anomalies in SDN. In NFV, it is helpful in grouping VNF instances, and NS is applied to group traffic of a homogeneous nature. Dimensionality reduction approaches such as PCA help in identifying key parameters within SDN, NFV, and NS. The unsupervised learning method and its application in context of SDN, NFV, and NS are summarized in the Table 1.4.

1.4.3 Reinforcement Learning

Reinforcement Learning is the method of learning an intelligent agent through an iterative process by interacting with its environment.

In the process of learning, agents take action to achieve specific goals, and according to the outcome of their actions, they receive feedback in the form of rewards and penalties. The agent makes the decision to maximize commutative reward. One common representation of reinforcement learning (Q-learning) is expressed as in equation (1.5);

(1.5)

Table 1.4 Unsupervised learning algorithm for achieving the NexGen objective.

Technology

Unsupervised learning algorithms

Use case

References

SDN

Clustering (e.g., K-Means)

Controller placement, resource optimization

[

62

,

63

]

PCA

Dimensionality Reduction,

[

65

]

Auto encoders

Anomaly Detection,

[

66

,

67

]

Density-Based Clustering (e.g., DBSCAN)

Dense region and network hotspot detection

[

68

]

NFV

Clustering (e.g., K-Means)

Traffic clustering

[

69

]

PCA

Complexity reduction in NFV

[

70

]

PCA

Traffic classification and provisioning

[

71

]

Auto encoders

Efficient network slice management

[

72

]

where Q(s,a) is the Q-value in state “s” for taking action “a” within the network. Learning rate (α) represents how quickly the network adopts the new information; R (s,a) represents the immediate rewards of action “a.” The discount factor (γ) is the balance factor between intermediate rewards and further consideration. Maxa Q(s’,a) denotes the maximum Q-value for the next state s’.

In the context of networking, optimal routing, resource allocation, and other dynamic processes require systems to adopt and learn from experience in the real-time networking environment. It helps to improve network performance and efficiency in response to dynamic network conditions. Reinforcement learning plays a significant role in facilitating adoptive and dynamic decision-making capabilities. Q-learning is employed to enhance adoptive routing optimization, network optimization, and traffic engineering in an SDN environment. For NFV, it facilitates the optimization and management of the NFV life cycle. In the context of network services, it plays a crucial role in achieving QoS-aware customization and E2E service provisioning. The reinforcement learning method and its application in context of modern paradigms are summarized in Table 1.5.

Table 1.5 Reinforcement learning algorithm for achieving the NexGen objective.

Technology

Reinforcement learning algorithms

Use case

References

SDN

Q-Learning

Adaptive routing optimization, load balancing

[

72

,

73

]

Deep Q Network (DQN)

Dynamic traffic prediction, adoptive network control and policy optimization

[

74

,

75

]

Policy Gradient Methods

Dynamic routing

[

77

]

Actor-Critic Methods

Adoptive network control, policy optimization and traffic engineering

[

78

,

79

]

NFV

Q-Learning

Resource optimization, adoptive SFC

[

80

,

81

]

Deep Q Network (DQN)

Adoptive SFC

[

82

]

Actor-Critic Methods

SFC, zero-touch networking

[

83

,

84

]

NS

Q-Learning

Dynamic resource allocation

[

85

]

Deep Q Network (DQN)

Service chaining and customized QoS-driven network orchestration

[

86

,

87

]

Actor-Critic Methods

Dynamic resource allocation

[

82

]

1.4.4 Deep Learning

Deep learning is considered a specialized ML model with multiple layers Deep Neural Network (DNN) to solve complex problems with large datasets. Deep learning can be expressed as a labeled dataset D, represented in equation (1.6);

(1.6)

where xn and yn represent the input feature and the corresponding output label, respectively. In the context of networking, the input feature could be device information or input traffic data, and the corresponding output label could be a classification or network event. While learning, multiple layers and their parameters are adjusted iteratively to minimize a defined loss function in equation (1.7);

(1.7)

Deep learning in networking represents a powerful tool for intelligent network automation. The traditional method may struggle to deal with manual feature engineering. Deep learning methods such as Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Recurrent Neural Networks (RNN) are well known for their excellent performance in image pattern recognition. In the context of networking, these technologies also offer advanced capabilities for traffic pattern recognition, anomaly detection, predictive analysis, and intelligent resource management. The reinforcement learning method and its application in context of modern technologies are summarized in Table 1.6.

Table 1.6 A Deep learning algorithm for achieving the NexGen objective.

Technology

Deep learning algorithms

Use case

Reference

SDN

CNN

Anomaly detection and traffic pattern classification

[

87

,

88

]

RNN

Time-series data analysis and traffic prediction

[

89

,

90

]

LSTM

Anomaly detection based on temporal patterns, routing optimization and traffic prediction

[

91

,

92

]

GAN

Traffic generation and Anomaly testing

[

93

,

94

]

NFV

CNN

VNFs deployment service function chaining and Predictive VNF Auto-scaling

[

95

,

96

]

RNN

Predictive analysis, dynamic scaling and life cycle management

[

97

,

98

]

LSTM

Temporal dynamics analysis for VNF scaling and life cycle management

[

99

]

GAN

Synthetic data generation for NFV testing and training

[

100

,

101

]

NS

CNN

Image-based QoS provisioning in network slices

[

102

]

RNN

Dynamic QoS prediction based on temporal dependencies, predictive maintenance based on historical data

[

103

]

LSTM

Dynamic resource allocation for slices based on historical data and E2E service provisioning

[

104

]

GAN

Synthetic data creation for network slice testing and analysis

[

105

]

In summary, AI-driven approaches, including ML and DL based methods, play a significant role in shaping intelligent and automated modern networking.

1.5 Navigating Challenges and Solutions

In this section, identify the specific challenges and research solutions arising from implementing 5G systems.

1.5.1 Performance Issues in Network Structure

Performance isolation in a network structure deployed over a common underlying infrastructure can be a challenging task. When multiple network slices or services share the same physical infrastructure, ensuring that each slice meets its performance requirements without interfering with others is a complex endeavor [106, 107]. Performance issues in a network structure can arise from various factors, and addressing them is crucial for ensuring the efficient operation of the network. Some common performance issues in network structures include bandwidth limitations, network congestion, latency, packet loss, network security measures, outdated hardware, inefficient routing, network protocol issues, network topology limitations, monitoring, and analysis. Regular network assessments, proactive monitoring, and a strategic approach to network design and optimization are essential for mitigating and preventing performance issues in a network structure.

1.5.2 Management and Orchestration Issues

Management and Orchestration (MANO) in the context of network virtualization and cloud computing involve the coordination of various resources and services to ensure efficient and reliable network operations [108]. However, several challenges and issues can arise in the process. There are some key management and orchestration issues, such as interoperability, integration with legacy systems, scalability, orchestration complexity, security concerns, resource Optimization, lifecycle management, multi-domain orchestration, service assurance, and vendor lock-in [109]. These management and orchestration issues require a holistic approach involving collaboration between industry stakeholders and the fulfillment of standards. Moreover, the continuous evolution of MANO technologies to meet the demands of dynamic and complex network environments.

1.5.3 Security and Privacy

The adoption of open interfaces and programmability in softwarized networks indeed introduces new potential attack vectors that need to be addressed with a robust security framework. The multi-level security framework mentioned should encompass various aspects to ensure the integrity, confidentiality, and availability of the network [110]. The security and privacy concerns arising from 5G-network infrastructure are the major barrier to adopting multi-tenancy approaches [111].