Adaptive video streaming architecture leveraging QoE forecasting and Content Steering on the Edge–Cloud Continuum

The demand for video streaming continues to grow, with HTTP Adaptive Streaming (HAS) dominating the Internet traffic. As a result, research interest is increasing in managing network congestion and operational costs while maintaining the user Quality of Experience (QoE). This work presents an adaptive video streaming architecture designed for scaling containers across the Edge-Cloud according to demand fluctuations. We highlight three main contributions: (i) The proposed architecture integrates the Content Steering Service (CSS) to select a suited endpoint during video sessions dynamically; (ii) Through continuous real-time monitoring, the system ensures seamless up-to-date information about the video playback, network, and node changes. iii) A deployed Machine Learning (ML) model is capable of forecasting users’ QoE, taking into account the prediction horizon, to proactively reconfigure the system before congestion occurs. In addition, the emulated network scenario, covering core, edge, and last-mile layers, demonstrates the system’s effectiveness. The results indicate significant improvements in the elimination of manual network configuration, highlighting the potential of edge-cloud integration for video streaming services. The ML model application demonstrates significant QoE maintenance, with up to 48% reduction in Service Level Objective (SLO) violations compared to a reactive approach while avoiding over-provision during low-traffic periods. • This architecture tackles the inherent challenges of scaling video services at the Edge, which considers the location-aware coverage of the Edge in multiple regions. This architecture scales video delivery to the Edge during traffic spikes while relying on cloud resources during lower demand periods. • An ML methodology is defined to deploy models within the architecture. To demonstrate its effectiveness, the methodology is evaluated in an emulated scenario. The research employs a LSTM predictive model to forecast QoE. • The architecture integrates continuous monitoring for containerized services in real time. This adaptability is an important feature in the Architecture component’s performance for dynamic network environments, ensuring seamless video playback.