Titre : | Deep Reinforcement Learning Based Approach for Service Function Chain Deployment in 5G Networks |
Auteurs : | Nour Elimane Elbey, Auteur ; Soheyb Ayad, Directeur de thèse |
Type de document : | Monographie imprimée |
Editeur : | Biskra [Algérie] : Faculté des Sciences Exactes et des Sciences de la Nature et de la Vie, Université Mohamed Khider, 2022 |
Format : | 1 vol. (79 p.) |
Langues: | Anglais |
Mots-clés: | 5G networks, Software Defined Network (SDN), Network Function Virtualisation (NFV), Service Function Chain (SFC), Deep Reinforcement Learning (DRL) |
Résumé : |
5G networks are intended to simultaneously support a wide range of applications with a high variety of requirements which brings a diversity of use cases for mobile networks and a growing number of demands. The development of 5G relies on new techniques such as Software Defined Networks (SDN), Network Function Virtualisation (NFV) and Service Function Chain (SFC) technologies. SDN allows the separation of control and data planes. NFV decouples network functions from the hardware that performs them through virtualisation. SFC is a popular service paradigm that has been proposed to derive maximum benefits from both NFV and SDN in 5G networks. The conventional approaches to service and infrastructure management, which require complete and perfect knowledge of the system, are ineffcient. In this context, Deep Reinforcement Learning (DRL), which marked a success in solving complicated control and decision-making problems by allowing network entities to learn, build knowledge, and make optimal decisions independently, inspired us to suggest this study, where we focus on optimising the resource utilisation of the fifth generation networks. We propose a Deep Q-learning (DQN) based DRL approach for optimal SFC deployment in 5G networks. The objective of the proposed approach is to deploy SFCs automatically and support the heterogeneity of SFC network requirements. Experimental results show that the approach proposed can achieve superior performance in solving SFC deployment problem, where the average return are 15.28% compared with the Q-learning based Reinforcement Learning (RL) approach 7.4%. |
Sommaire : |
Acknowledgement I Abstract II List of Abbreviation VII List of Figures IX List of Tables XI List of Algorithms XII List of listings XIII General introduction 1 1 5G Network on General 4 1.1 Introduction .. 4 1.2 Mobile Network Evolution 5 1.3 5G Network . 5 1.3.1 5G Architecture . 6 1.3.2 5G Key Performance Indicators (KPI) 7 1.3.3 5G Enabling Technologies 9 1.4 Network Function Virtualization (NFV) . 12 1.4.1 NFV architecture .. 12 1.5 Software Defined Network (SDN) 14 1.5.1 SDN architecture 14 1.6 Service Function Chain (SFC) 16 1.6.1 Static service function chaining 17 1.6.2 Dynamic service function chaining 17 1.6.3 Service function chaining architecture 18 IVCONTENTS 1.6.4 SDN/NFV architecture for SFC deployment 20 1.7 Conclusion 21 2 Machine Learning and Networking 22 2.1 Introduction 22 2.2 Machine Learning 22 2.2.1 Machine Learning Types 23 2.2.2 Difference between Machine Learning types 26 2.3 Deep Reinforcement Learning . 27 2.3.1 Policy and Value function 27 2.3.2 Exploration vs Exploitation 28 2.3.3 Deep Q-learning 29 2.4 Machine Learning in Networking 31 2.4.1 Special Considerations to deploy ML in Networking 34 2.4.2 Machine Learning for 5G Network 34 2.5 Related Works 36 2.5.1 DRL for NFV-based Service Function Chaining in Multi-Service Networks 36 2.5.2 Online Service Function Chain Deployment with DRL in 5G networks 36 2.5.3 Adaptive Online SFC Deployment with Deep Reinforcement Learning 37 2.5.4 Q-Learning based SFC deployment on Edge Computing Environment 37 2.5.5 RL based QoS/QoE-aware Service Function Chaining in Software-Driven 5G Slices 37 2.5.6 Discussion 38 2.6 Conclusion 39 3 Design 40 3.1 Introduction 40 3.2 System Modeling 40 3.2.1 Network Definition 40 3.2.2 Service Function Chain Request 40 3.2.3 Problem Formulation 41 3.2.4 Markov Decision Process (MDP) for SFC deployment 41 3.3 General Architecture . 42 3.4 Detailed Architecture 43 3.4.1 Source and Destination nodes 44 3.4.2 Network Function Virtualization 44 VCONTENTS 3.4.3 Software Define Network 45 3.4.4 Network topology 45 3.4.5 DRL model architecture 46 3.5 Reinforcement Learning model architecture 55 3.5.1 RL based approach for SFC deployment algorithm 55 3.5.2 RL based approach for SFC deployment process 57 3.6 Conclusion . 58 4 Experimental study and results 59 4.1 Introduction 59 4.2 Development tools 59 4.2.1 IDE (integrated development environment) 60 4.2.2 Programming language 60 4.2.3 libraries 60 4.3 Implementation 61 4.3.1 Environment 61 4.3.2 Deep Q-learning Agent 63 4.3.3 Q-learning Agent 67 4.4 Results . 69 4.4.1 DQN models evaluation results 70 4.4.2 DQN vs Q-learning comparison results 73 4.5 Final discussion 75 4.6 Conclusion 76 General conclusion 77 Bibliography 79 |
Disponibilité (1)
Cote | Support | Localisation | Statut |
---|---|---|---|
MINF/710 | Mémoire master | bibliothèque sciences exactes | Consultable |