30 April 2025
09:00 Doctoral defense IC3 Auditorium
Topic on
An Architecture for Adaptive Video Streaming: QoE Forecasting, Content Steering, and Scalable Resource Management through Edge-Cloud Computing
Student
Eduardo de Souza Gama
Advisor / Teacher
Luiz Fernando Bittencourt - Co-advisor: Roger Kreutz Immich
Brief summary
Video streaming services represent the majority of Internet traffic. The increasing use of Internet-connected devices and the emergence of diverse content providers have driven this demand. This presents a major challenge for network operators in ensuring Quality of Experience (QoE) metrics, which is already complicated by the adaptive behavior of DASH mechanisms. Motivated by this, this thesis presents a learning-based architecture for adaptive video streaming at the edge and cloud network, addressing challenges in QoE, scalability, and resource management by integrating HTTP Adaptive Streaming (HAS) with Content Steering Services (CSS). The proposed system dynamically directs user requests to optimize video delivery transparently to HAS clients. First, we design, analyze, and evaluate an architecture that leverages multi-tier cloud and edge environments by dynamically allocating video services across multiple edge and cloud nodes. Challenges such as edge node selection, resource scalability, and dynamic user mobility, which impact network efficiency and user satisfaction, are discussed. Experimental evaluations demonstrate the need for such environments to have an architecture capable of adapting to changing network demands in real time. This impacts the number of interruptions in video playback, bitrate stability, and QoE in different scenarios, including environments with mobile users and dynamic load conditions. We then incorporate into the developed adaptive video streaming architecture an architecture that integrates edge and cloud computing, CSS, and machine learning (ML) techniques to optimize video delivery while ensuring scalability and resource efficiency. The proposed system employs HAS technologies and incorporates a Planner Service capable of allocating resources and predicting Service Level Objective (SLO) violations. By dynamically routing user requests, the architecture achieves significant reductions in latency, improved playback stability, and improved cache utilization. Experimental results demonstrate significant improvements in QoE and system performance compared to traditional approaches, highlighting reduced latency, improved cache efficiency, and continuous adaptation to dynamic network conditions. This work contributes novel algorithms and prediction mechanisms to the adaptive video streaming domain, emphasizing scalability, responsiveness, and efficient resource utilization in modern HAS environments.
Examination Board
Headlines:
| Luiz Fernando Bittencourt | IC / UNICAMP |
| Kelvin Lopes Dias | CIn / UFPE |
| Denis Lima do Rosário | ICEN / UFPA |
| Juliana Freitag Borin | IC / UNICAMP |
| Rodolfo Jardim de Azevedo | IC / UNICAMP |
Substitutes:
| Sandro Rigo | IC / UNICAMP |
| Geraldo Pereira Rocha Filho | DCET/UESB |
| Fabio Luciano Verdi | DComp / UFSCar |