Real Time ML-Based QoE Adaptive Approach in SDN Context for HTTP Video Services

Abstract

Due to the high dynamism of network conditions, operators and service providers are facing the challenge of providing satisfactory user experience during a real-time video streaming session where clients are often suffering from frequent interruptions and significant visual quality degradation. Video parameters such as playback quality, rate switching amplitude/frequency, occupancy, overflow/underflow buffer are the main key factors responsible for affecting the user experience’s quality. Recently, adaptive streaming protocols over HTTP have become widely adopted for providing continuous video streaming services to users with their different heterogeneous devices under dynamic network conditions. In this paper, we leverage the emerging paradigm of software defined networking SDN. Our contribution consists in developing some scenarios on SDN helping to adapt video streaming to the network state. The current work proposes to experience ML algorithms in order to predict user QoE over SDN networks. We present an approach that collects MOS score from users under varying network parameters as well as objective parameters such as SSIM, VQM and PSNR. The MOS scores are collected by playing videos to actual users in an SDN environment. We design an architecture which could use the measured MOS values under varying network conditions to predict the expected MOS based on machine learning algorithms. This work provides an outlook of experiments done for demonstration, by describing SDN environment deployment, detailing the realized scenarios and finally giving the results and values. We highlight, at the end of this paper, the perspectives of our proposition.