검색 상세

Machine Learning-Based QoS Enhancement for SD-IoT Networks

초록/요약

The Quality of Service (QoS) requirements of Internet of Things (IoT) networks vary significantly across different applications. Integrating IoT with Software Defined Networks (SDN) creates a more flexible and dynamic Software Defined-Internet of Things (SD-IoT) network, capable of addressing diverse QoS demands. However, achieving QoS and enhancing performance in such networks necessitates the use of specialized techniques. This dissertation explores three distinct methodologies aimed at improving QoS in SD-IoT networks. In SDN, routing decisions are managed by the SDN controller, which identifies optimal paths for incoming traffic flows. However, conventional routing techniques often fall short in meeting the unique QoS demands of various types of flows. Traffic classification techniques address this gap by categorizing incoming flows into distinct classes, enabling the controller to tailor routing strategies to specific flow require- ments. Additionally, challenges such as link reliability and delay further complicate QoS provisioning. Predictive analysis techniques can forecast these metrics, allowing the controller to make informed decisions that align with the network’s current and an- ticipated conditions. By incorporating predictive insights into routing and traffic management, the SD-IoT network can dynamically adapt to varying demands and ensure enhanced QoS for diverse applications. Firstly, we investigate techniques aimed at enhancing the QoS of SD-IoT net- works. Building on these insights, we propose a novel hybrid framework to further improve QoS [1]. Our approach introduces a dynamic routing module that integrates reinforcement learning (RL) with classification and reliability prediction modules. The classification module categorizes incoming flows into distinct classes based on their QoS requirements, while the reliability prediction module forecasts the reliabil- ity of each network link. The RL framework utilizes these metrics to compute optimal paths, ensuring reliable and efficient data transmission. The proposed framework is evaluated using various real-world internet topologies and assessed against key QoS metrics, including delay, throughput, packet loss ratio, and jitter. This comprehen- sive evaluation demonstrates the framework’s effectiveness in improving the overall performance of SD-IoT networks. Secondly, we focus on the classification module, which addresses the challenges posed by imbalanced classes of IoT applications and the continuous emergence of new applications. To effectively manage these challenges, an automatic classification module capable of handling both class imbalance and adaptability to new applications is essential. To this end, we propose a novel classification algorithm, AL-CSXGB. This algorithm is specifically designed to address the imbalance problem in traffic classification. By incorporating active learning and cost-sensitive techniques, our approach not only enhances classification accuracy but also adapts to evolving traffic patterns [2]. Evaluation results demonstrate that the proposed AL-CSXGB algorithm significantly outperforms other classification approaches, making it a robust solution for traffic classification in dynamic SD-IoT environments. Thirdly, we focus on analyzing QoS improvements through delay prediction in SD-IoT networks. Frequent interactions among IoT devices, SDN switches, and con- trollers contribute significantly to network delay, making delay prediction a critical aspect of enhancing QoS. In this work, we investigate TCP delay by first generating TCP flows in an emulated Mininet environment to create a comprehensive dataset that includes delay metrics. To accurately predict TCP delay, we employ an en- semble learning algorithm designed to combine the strengths of multiple regression models. This approach effectively captures both linear and non-linear delay patterns, enabling more precise predictions. Finally, we compare the proposed delay predic- tion algorithm with baseline algorithms and evaluate its performance across various real-world network topologies. The results are analyzed in terms of QoS metrics such as delay, throughput, and jitter, demonstrating the effectiveness of our approach in improving QoS in SD-IoT networks. Keywords: Software defined-Internet of Things Networks, Quality of Service, Routing, Traffic classification, Delay prediction, Active learning, Machine learning.

more

목차

I. Introduction 1
1.1 Motivation and Objectives 1
1.2 Research Contributions 3
1.2.1 QoS Enhancement via Intelligent Path Prediction in SD-IoT . 4
1.2.2 Effective Imbalanced Traffic Classification in SD-IoT Using XG- Boost and Active Learning 4
1.2.3 Ensemble Learning with Weighted Fuzzy Logic for TCP Delay Prediction in SD-IoT 5
1.3 Organization of the Dissertation 6
II. Background 7
2.1 Software Defined-Internet of Things 7
2.1.1 SD-IoT Architecture 9
2.1.2 IoT Infrastructure Plane 9
2.1.3 Data Plane 11
2.1.4 Southbound APIs 11
2.1.5 Northbound APIs 12
2.1.6 Application plane 13
2.1.7 control plane 13
2.2 SD-IoT QoS provisioning methods 18
2.2.1 Routing Optimization 18
2.2.2 Predictive Analytics 19
2.2.3 Traffic Classification 20
2.2.4 Hybrid methods 22
III. QoS Enhancement via Intelligent Path Prediction in SD-IoT 24
3.1 Introduction 24
3.2 Motivation and problem formulation 27
3.3 Proposed work and research contributions 30
3.3.1 IoT Device Plane 30
3.3.2 Data Plane with Intelligent Network Information Monitoring (INIM) 31
3.3.3 Control Plane 33
3.3.4 Application Plane 35
3.3.5 Reliability prediction using SVR 35
3.3.6 RL based routing leveraging reliability and path prediction 38
3.4 Experimental setup and results 45
3.4.1 Experimental setup 45
3.4.2 Results 48
3.5 Summary 54
IV. Effective Imbalanced Traffic Classification in SD-IoT Using XGBoost and Active Learning 56
4.1 Introduction 56
4.2 Problem statement and contributions 63
4.3 Proposed approach 65
4.3.1 Architecture of AL-CSXGB 65
4.3.2 Data preprocessing for AL-CSXGB 67
4.3.3 XGBoost framework 70
4.3.4 Cost sensitive XGB (CSXGB) 73
4.3.5 AL-CSXGB 74
4.4 Experimental setup and performance evaluation 78
4.4.1 Experimental Setup 78
4.4.2 Performance evaluation 79
4.5 Summary 97
4.6 Challenges 98
V. Ensemble Learning with Weighted Fuzzy Logic for TCP Delay Prediction in SD-IoT 100
5.1 Introduction 100
5.2 Motivation and research contributions 102
5.3 Problem Formulation 105
5.4 Proposed Framework 107
5.4.1 System Model 108
5.4.2 Dataset Creation 110
5.4.3 Data Pre-processing 112
5.4.4 Base Learners 115
5.4.5 Fuzzy Weighted Ensemble Learning 117
5.5 Experimental Setup and Performance evaluation 121
5.5.1 Experimental Environment 121
5.5.2 Evaluation Metrics 121
5.6 Performance evaluation 124
5.6.1 Summary 133
VI. Conclusion and Future work 135
6.1 Conclusion 135
6.2 Future work 136
References 138

more
반출 Meta View 목록