Software-Defined Networks: Architecture for Extended SDN Applications and Resource Optimization in Cloud Data Centers

Publication Type:
Issue Date:
Full metadata record
Virtualization is the main mechanism to share resources to many customers by creating virtual resources on the common physical resources. The challenge is to search for an optimal resource allocation mechanism that maximizes the capacity of the virtual resources. Network virtualization needs a new virtual network embedding (VNE) mechanism that focuses concurrently on control congestion, cost saving, energy saving; a link embedding mechanism needs to select actively based on multiple objectives the physical link resources, network slicing requires a new resource allocation mechanism that satisfies latency constraints of 5G mobile system. This research investigated and developed solutions for resource request delivery, and optimal resource allocation in network virtualization and 5G core network slicing applying SDN technology. In the research, firstly, the three-tier architecture applying micro-service architecture for extended SDN application is presented to facilitate the flexibility, in which new services are created or composed, existing services are reused. The evaluation is the prototype of the Dynamic resource allocation using the proposed architecture. Secondly, the multiple-objective VNE that focuses on congestion avoidance, energy saving and cost saving (CEVNE) is presented. The novelty lies in the CEVNE mathematical model for multiple-objective optimization problems, and its nodes and link embedding algorithms. The evaluation showed that CEVNE outperformed The-State-Of-The-Art in acceptance ratio in the challenged, near-congestion scenarios. Thirdly, the architecture to realize virtual link mapping in CEVNE is presented. The novelty is in the SDN-based heuristic algorithm, and the applying of the architecture for extended SDN applications. The research results in the realization of the active virtual link embedding process that focuses on multi-objective concurrently. The evaluation showed that the solution outperformed the traditional link mapping in all three objectives. Fourthly, the mathematical model of the resource allocation optimization in latency-aware 5G core network slicing is presented. The novelties lie in the satisfaction of different latency requirements of 5G applications: eMBB, uRLLC, and mMTC, and the solution strategy to linearize, convex-relax and decompose the program into sub-problems. The evaluation shows that the solution outperformed the The-State-Of-The-Art in resource allocation, execution time, latency satisfaction and the arrival rates. In this thesis, the resource optimization problem and the architecture for extended SDN applications have been studied comprehensively. The results of this thesis can readily be applied to 5G vertical applications where resource optimization and network routing problems exist naturally in multiple domains and require software defined networking logically centralized control architecture for efficient and dynamic solutions.
Please use this identifier to cite or link to this item: