CPU Resource Allocation in SDN, SDR and MEC Integrated Networks

Abstract

Software-Defined Networking (SDN) and Software-Defined Radio (SDR) are two key oftwaredefined technologies for enabling flexible and open wired and wireless connections in nextgeneration etworks. SDN separates the control plane from the data plane, allowing for centralized, programmable control over network functions, while SDR uses software to implement radio functions, providing daptability in wireless communication systems. Network Function Virtualization (NFV) virtualizes network functions, making it possible to integrate SDN and SDR within the same cloud infrastructure, thus reducing the reliance on specialized hardware and enhancing scalability. Researchers have recently explored use cases for such integration, with cloud-based platforms like Mobile Edge Cloud (MEC) being one of these promising applications. MEC brings computation closer to end users by extending cloud capabilities to the edge, which helps meet the low-latency requirements of modern applications such as augmented reality, autonomous driving, and real-time data analytics. Our investigation highlights that the cloud infrastructure used for SDN and SDR can also support MEC applications, offering an opportunity to integrate SDN, SDR, and MEC into a unified cloud infrastructure. Such integration allows for more dynamic and flexible resource utilization, leading to better performance, cost savings, and energy efficiency. In this thesis, we propose an integrated network architecture that leverages this integration, which, to the best of our knowledge, is a novel contribution to this research area. Moreover, recent literature on SDN and SDR has primarily focused on the performance of communication aspects, such as bandwidth, delay, and packet loss, to meet the low-latency and high-bandwidth requirements of 5G and beyond. However, these studies often overlook computational resource requirements and related power consumption due to limited real data on computational resource usage in cloud environments and a lack of experimental testbeds. This thesis addresses this gap by focusing on CPU resource allocation and energy-saving strategies for SDN, SDR, and MEC integration. We designed and conducted experiments using the Trinity College Dublin OpenIreland testbed with real SDN and SDR equipment to evaluate CPU and power consumption. Our results provide insights into computational resource savings when applying our approaches to an integrated MEC environment, examining the correlation between network bandwidth and CPU consumption for SDN and SDR. The findings are further applied to MEC use cases and experiments. Additionally, we investigate algorithmic solutions for resource allocation in a shared SDN, SDR, and MEC structure. Existing literature mainly focuses on centralized control algorithms, where a central controller manages resource allocation for all elements in an MEC network. There is, therefore, a gap in distributed resource sharing and self-organization. This thesis focuses on distributed solutions for computational resource sharing through self-organization. Our approach shares spare resources between users to guarantee their satisfaction, employing a self-organization-based method for CPU resource sharing while considering users' personality traits. The resource-sharing steps are game-theory-based and utilize distributed solutions. Furthermore, we simulated real MEC use cases to demonstrate the benefits of our proposed algorithms, showing improvements in resource allocation and power savings over baseline algorithms. In summary, this thesis covers the theoretical, simulation, and experimental investigation of CPU resource allocation for SDN, SDR, and MEC in a shared cloud-based environment. Our key contributions beyond the state-of-the-art are as follows: 1) We propose a softwarized network architecture that integrates the computation and processing of SDN, SDR, and MEC within a unified cloud-based infrastructure, allowing for enhanced flexibility and dynamic resource allocation. 2) We optimize CPU and power consumption for this integrated infrastructure, supported by both testbed experiments and simulation results, demonstrating significant improvements in resource utilization and energy efficiency compared to existing approaches.