Multimedia Traffic Management over Software-Defined Networking

In the modern landscape of multimedia traffic transmission, achieving efficient network management and ensuring quality of service for diverse video services become a challenging task. Meanwhile, the introduction of virtualisation through the integration of software defined networking allows for the effective control of traffic and remote management of network devices. Nevertheless, the growing popularity and diversity of video-based multimedia applications, coupled with the expanding user base, have created significant challenges for the underlying networks. This surge in multimedia content applications has put additional pressure on the networks, hence an intelligent traffic management solution is imperative to ensure the provision of quality of service. Machine learning techniques are also becoming increasingly essential for intelligent modern network management. To achieve automatic multimedia traffic management with the insurance of quality of service and end-user quality of experience, the integration of software-defined networking with machine learning techniques is required to provide better solutions. The work in this thesis intends to design and develop an intelligent-based framework for multimedia traffic management over software-defined networking. First, the thesis addresses the challenge associated with the individual classification of multimedia traffic: from the quality of service perspective. An experiment is conducted to capture video traffic from three video streaming services, Amazon Prime Video, Netflix and YouTube. The reason behind using these services is to realise their quality of service priority requirements for restoration, considering the subscription levels of each service. A multi-layer feedforward neural network model is constructed and employed for the classification experiment, resulting in an impressive overall accuracy of 99.1%. Subsequently, another experiment is undertaken to collect data from the same services. This data is utilised to assess the model’s performance in classifying unseen data instances, achieving a high accuracy of 98%. Furthermore, an intelligent-based multimedia traffic routing framework that exploits the integration of a reinforcement learning technique with software-defined networking to explore, learn and find potential routes for video streaming traffic is introduced. Realtime dynamic adaptive streaming over HTTP is utilised to evaluate the performance of the proposed solution. The goal of this strategy is to enable service providers to route video traffic through paths with high bandwidth while avoiding selecting paths with high delays, jitter and packet loss rates, with the objective of providing better QoE for the end-users. Simulation results with different realistic network topologies and under various traffic loads demonstrate the proposed scheme’s effectiveness in providing a better end-user viewing experience, and higher throughput compared to benchmark algorithms. The thesis further proposes a new mechanism to enable multimedia traffic management over softwarsied networks. The scheme involves a fine-grained classification of various multimedia applications and reinforcement learning for route selection over software-defined networking. The solution is driven by the idea that video applications may have varying quality of service and network resource requirements, while video flows within the same category typically have similar quality of service requirements. Three services are defined including live streaming, video-on-demand streaming, and download streaming. Each service exhibits unique performance characteristics, technical features, and quality of service demands. The goal is to highlight the importance of addressing their specific requirements to optimise the quality of service delivery effectively, leading to enhanced clients’ viewing experience.