제주대학교 Repository

Ensemble-based Prediction Scheme for Resource Utilization in IBN-enabled Network Slice Lifecycle Management

Metadata Downloads
Abstract
5G networks come up with many innovative features compared to legacy networks, such as network slicing that envisioned a wide variety of services from different customers, network operators, and industrial verticals. Network slicing is the partitioning of a physical network into multiple logical isolated networks. It ensures dedicated and isolated resources to each of the services. The autonomous orchestration and management of end-to-end (e2e) network slicing is critical due to the complex network configuration for the underlying infrastructure. On the other side, data analytics seems promising to manage and control the underlying network resources proactively. So, Network Data Analytics Function (NWDAF) has been introduced in 5G service-based architecture (SBA), which enables network operators to use various Artificial Intelligence (AI) and Machine learning (ML) techniques. These ML models are trained on historical network data collected from multiple domains such as core, RAN, and edge. It allows network operators to implement their own or third-party ML mechanisms. More specifically, the proactive management of cloud resources is still a challenging task. Therefore, this thesis primarily focuses on e2e network slice lifecycle management and AI and ML-based network data analytics mechanisms for proactive management of network resources.
An Intent-based Networking (IBN) mechanism has been developed to automatically control, orchestrate, and manage e2e network slicing. It follows a closed-loop approach for the network slice lifecycle management (LCM). The results achieved through the proposed mechanism show satisfactory performance. Moreover, motivated by NWDAF, a data analytics mechanism has been integrated with the IBN platform to achieve proactive resource updates and assurance. This network data analytics mechanism uses novel hybrid ensemble learning (EL) algorithms for network resource utilization prediction and anomaly detection and mitigation. With the help of results, it can be observed that the developed mechanism outperformed the considered algorithms. In addition, ML models assist the IBN platform in updating and managing the network resources proactively.
Author(s)
Abbas, Khizar
Issued Date
2022
Awarded Date
2022. 2
Type
Dissertation
URI
https://dcoll.jejunu.ac.kr/common/orgView/000000010601
Alternative Author(s)
아바스 키자르
Affiliation
제주대학교 대학원
Department
대학원 컴퓨터공학과
Advisor
송왕철
Table Of Contents
INTRODUCTION 1
1.1. Research Problems and Objectives 7
1.2. Thesis Organization 9
RELATED WORK 11
2.1. E2E Network Slice Orchestration and Management 11
2.2. Standardization and Industrial Progress towards Network Automation 17
2.2.1. Standardized Bodies for Network Automation 17
2.2.2. Industrial Progress and Solutions for Automating the Network 23
2.3. AI and ML Approaches for Network resource Utilization Prediction and Anomaly Detection and Mitigation 27
2.3.1. Ensemble Learning Approaches 32
DESIGN AND ARCHITECTURE OF ENSEMBLE LEARNING-BASED NETWORK RESOURCE UTILIZATION PREDICTION FOR IBN-ENABLED SLICE LCM 34
3.1. Introduction 34
3.2. Intent-based Networking (IBN) platform for e2e Network Slice lifecycle Management 36
3.2.1. Slice Instantiation or Commissioning 37
3.2.2. Slice Activation 38
3.2.2.1. NFV- Orchestrator OSM for the Deployment of core VNFs 39
3.2.2.2. RAN Controller 40
3.2.3. Slice Run-time Monitoring 42
3.2.4. Slice Deactivation or Decommissioning 44
3.2.5. Decision Engine 46
3.3. Network Data Analytics Function (NWDAF) with IBN for Proactive Update and Assurance 47
3.3.1. Dataset Preprocessing 49
3.3.2. Proposed Hybrid Stacking Ensemble Learning (HSTEL) Model for Network Resource Utilization Prediction 53
3.3.2.1. Gradient Boosting Machine (GBM) 55
3.3.2.2. Gradient Boosting Model (XGBoost) 55
3.3.2.3. Catboost Model 57
3.3.3. Hybrid Model for Anomaly Detection 60
3.3.3.1. Random Forest (RF) 61
3.3.3.2. Dataset Information 62
EXPERIMENTAL RESULTS AND DISCUSSION 66
4.1. Results of E2E Network Slicing through IBN System 66
4.1.1. Experimental Testbed Details 67
4.1.2. Results and Discussion of Network Slicing 70
4.2. Results of HSTEL Model for Network Resource Utilization Prediction 75
4.2.1. Performance Metrics for Model Evaluation 75
4.2.2. HSTEL Model Prediction Results 77
4.3. Results of Anamoly Detection through Hybrid Model 87
CONCLUSIONS 90
BIBLIOGRAPHY 93
Degree
Doctor
Publisher
제주대학교 대학원
Appears in Collections:
General Graduate School > Computer Engineering
Authorize & License
  • AuthorizeOpen
Files in This Item:

Items in Repository are protected by copyright, with all rights reserved, unless otherwise indicated.