Citation

Richard, Etengu (2023) AI-enhanced traffic prediction for energy-efficient traffic engineering in hybrid SDNs. Other thesis, Multimedia University.

Full text not available from this repository.

Abstract

Recently, the rapid growth in data traffic volumes has overstretched the capacity of existing core infrastructure in modern communication networks, compromising network management and control to deliver high capacity multimedia services. Hybrid softwaredefined networking/open shortest path first (SDN/OSPF) networks represent a transitory network architecture for large-sized Internet service providers to cope with the explosive growth in traffic volumes. Apart from quality-of-service assurance, improving energy efficiency is one of the key goals of SDN/OSPF networks from environmental and economic perspectives. To date, various adaptive link rate energy-aware routing (ALREAR) traffic engineering (TE) methods have been investigated to improve the energy efficiency of such hybrid networks. However, these methods are either traffic-unaware or slow, resulting in underperformance. This study investigates an intelligent optimisation framework that uses deep learning (DL) and a bio-inspired genetic algorithm (GA) for data traffic prediction and ALR-EAR in hybrid SDN/OSPF networks. The ALR-EAR problem is solved using a metaheuristic-based bio-inspired GA, formulated as a multiobjective integer linear programming (MOILP). To guarantee rapid and intelligent generation of energy-efficient routing decisions, various tasks are performed. First, we develop a novel supervised deep learning framework that enhances traffic prediction in transitional hybrid SDN/OSPF networks by combining multiple dimensionality reduction techniques and a multi-objective GA. Second, the proposed ML-ALR-GA-EAR method utilizes classical ALR-GA sleeping simulation-based hyper-tuning to optimize population initialisation and fitness evaluation function computation, facilitating intelligent training parameter generation. Third, the DL prediction module is designed and trained offline for routing performance optimisation, and the online ALR-GA-EAR module is used to select optimal hyperparameters for network energy consumption. The ALR-GA-EAR module is employed to adjust traffic demands and routes accordingly. Lastly, after being trained, the final model is deployed to generate near-optimal energy-efficient routes. The ML traffic prediction method, developed using a Mininet emulator on Ubuntu, outperforms traditional ML and DL frameworks in accuracy and power savings, demonstrating potential for improved operation and management of current and future SDN networks.

Item Type:	Thesis (Other)
Additional Information:	Call No.: TK5105.5833 .E84 2023
Uncontrolled Keywords:	Software-defined networking (SDN)
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5101-6720 Telecommunication. Including telegraphy, telephone, radio, radar, television
Divisions:	Faculty of Computing and Informatics (FCI)
Depositing User:	Ms Nurul Iqtiani Ahmad
Date Deposited:	29 Sep 2025 08:35
Last Modified:	29 Sep 2025 08:35
URII:	http://shdl.mmu.edu.my/id/eprint/14506

Downloads

Edit (login required)