Adaptive QoE monitoring architecture in SDN networks: Video streaming services case

Abstract

In data services application over the internet, the user perception and satisfaction can be assessed by Quality of Experience (QoE) metrics. As QoE depends on both the users' perception and the used service, they form end-to-end metrics. While network optimization has traditionally focused on optimizing network properties such as QoS, in this work, we focus on optimizing end-to-end QoE metrics and hope to deliver to the client a good QoE and monitor it on real time. We argue that end-user QoE is the measure that is relevant for network operators and service providers. In today's world, video streaming rose above all other types of traffic. In fact, providing this service with a high quality presents the most challenging task among the advancements in networking technologies. Researchers are trying to help creating a more efficient network where congestion, broadband limitations and skyrocketing number of users present ever-diminishing obstacles. When it comes to us, we present in this paper a machine learning approach combined with adaptive video delivery service in order to provide a better QoE for video streaming services. This solution will be established using an SDN architecture. We can justify this choice because we need a centralized architecture, where the totality of the network is known, to predict its status. First part of the paper deals with a brief introduction of QoE and mathematical tools helping to model it. A synthetic study is done for this purpose. Second part describes the SDN networks, to see QoE requirement and service architecture to make simple the simulation deployment phase. The third part of the paper expose our proposed architecture, it describes the hole modules still the Rating Web application, ML model for predicting MOS, adaptive QoE monitoring concept, until the architecture of simulated environment. This application proposes to collect network parameters and modify video metrics thanks to user estimated MOS, network parameters measured such as RTT, Jitter, bandwidth and delay and objective parameters such as VQM, PSNR and SSIM. We highlight at the end the future of our proposition.