Resumen de Identifying and diagnosing video streaming performance issues
On-line video streaming is an ever evolving ecosystem of services and technologies, where content providers are on a constant race to satisfy the users' demand for richer content and higher bitrate streams, updated set of features and cross-platform compatibility. At the same time, network operators are required to ensure that the requested video streams are delivered through the network with a satisfactory quality in accordance with the existing Service Level Agreements (SLA).
However, tracking and maintaining satisfactory video Quality of Experience (QoE) has become a greater challenge for operators than ever before. With the growing popularity of content engagement on handheld devices and over wireless connections, new points-of-failure have added to the list of failures that can affect the video quality. Moreover, the adoption of end-to-end encryption by major streaming services has rendered previously used QoE diagnosis methods obsolete.
In this thesis, we identify the current challenges in identifying and diagnosing video streaming issues and we propose novel approaches in order to address them. More specifically, the thesis initially presents methods and tools to identify a wide array of QoE problems and the severity with which they affect the users' experience. The next part of the thesis deals with the investigation of methods to locate under-performing parts of the network that lead to drop of the delivered quality of a service.
In this context, we propose a data-driven methodology for detecting the under performing areas of cellular network with sub-optimal Quality of Service (QoS) and video QoE. Moreover, we develop and evaluate a multi-vantage point framework that is capable of diagnosing the underlying faults that cause the disruption of the user's experience. The last part of this work, further explores the detection of network performance anomalies and introduces a novel method for detecting such issues using contextual information. This approach provides higher accuracy when detecting network faults in the presence of high variation and can benefit providers to perform early detection of anomalies before they result in QoE issues.
