Analysis and optimization of streaming media QOE over QUIC enabled modern browsers over varied network bandwidth

Abstract

Today, global Internet traffic is dominated by video streaming, prompting service providers to adopt emerging technologies that enhance users’ Quality of Experience (QoE) while sustaining operational efficiency. In this landscape, Google introduced QUIC in 2013, a transport-layer protocol engineered to improve web performance and video delivery, especially in environments characterized by high packet loss, variable throughput, and constrained bandwidth. At the same time, the rapid deployment of high-speed access technologies such as WiFi 6/7 and 5G-capable of delivering hundreds of megabits to multiple gigabits per second is reshaping the Internet delivery ecosystem. These advancements create a critical need to evaluate whether QUIC can consistently meet the demands of both constrained and high-speed networks. Although QUIC promises improved performance, its behavior in real-world, heterogeneous environments, across browsers and network conditions is still underexplored, which directly impacts video streaming’s QoE. Recognizing these limitations and developments, this thesis primarily focuses on understanding how QUIC is implemented within modern web browsers and how these implementation choices influence video streaming performance. To enable this investigation, we develop a specialized measurement tool and collect a large-scale dataset across diverse network conditions, allowing us to systematically analyze browser-specific QUIC behavior in real-world streaming scenarios. To this end, the thesis contributes a measurement tool and a large-scale dataset for analyzing QUIC’s real-world behavior along with the design of an intelligent, browser-integrated decision system that leverages network parameters to both dynamically control connection racing and select between QUIC and TCP based on prevailing network conditions, thereby improving video streaming performance. Specifically, this thesis investigates two fundamental questions: • How do browser-specific implementations of QUIC, in particular their protocol selection mechanisms such as connection racing, impact video streaming performance under varied network conditions? • Given QUIC’s performance limitations under certain network conditions, is it viable for browsers to rely on a single default transport protocol? Is there a need for an adaptive protocol selection mechanism? To answer these questions, we begin by examining the current browser implementation, where most modern browsers support HTTP/3 and rely on QUIC, which runs on top of UDP as the default transport protocol. Many Internet middleboxes either block or rate-limit UDP traffic, which can hinder QUIC connectivity. To ensure reliability and backward compatibility, browsers use a strategy known as connection racing, where both QUIC and TCP connections are initiated in parallel. The protocol that completes the handshake first is selected for subsequent communication. This raises a critical question about the conditions under which QUIC truly provides advantages over TCP, and when TCP may still be the preferable option. To study this in detail, the thesis develops an HTTP/3-compatible browser-based dataset collection tool, H3B, which captures both application-level and network-level logs during YouTube streaming. Using H3B, we conduct a large-scale measurement study across popular browsers (Chrome/Chromium and Firefox), covering 6013 YouTube sessions and 5474 hours of playback. Our analysis reveals that YouTube QoE often degrades due to repeated connection racing events. We modify the Chromium source code to disable connection racing, and we observe a significant improvement in QoE. Building upon this finding, we design two complementary solutions: (i) an adaptive mechanism to decide when connection racing should be enabled, and (ii) a server-side emulation framework to mitigate protocol switching impact. Instrumenting the Chromium codebase further shows that selectively disabling or controlling connection racing substantially enhances QoE by 60% compared to the baseline implementation. While this addresses the challenges arising from QUIC’s connectivity issues and the instability introduced by racing, our experiments, along with evidence from prior studies, show that it continues to exhibit performance limitations in high-speed networks. This insight naturally leads to the second question: Is it appropriate for browsers to rely on a single default transport protocol across all network scenarios? Hence, In this thesis, to address this, we propose IntelSwitch a novel client-side framework that enables adaptive switching between TCP and QUIC for video streaming. It does not require any server-side modifications. IntelSwitch continuously monitors network and QoE-related metrics through a DASH player and employs a reinforcement learning (RL) based decision engine to dynamically select the most suitable transport protocol based on real-time network and host conditions. The framework comprises two components: (i) modifications to the DASH player and Chromium browser to support seamless protocol switching during playback, and (ii) an RL-based decision module that determines the optimal protocol under varying network and system states. We implement IntelSwitch by extending Dash.js and the Chromium browser with protocol-aware playback. We found that adaptive transport protocol switching within the browser leads to an average improvement of up to 114% over choosing only QUIC as the transport protocol across diverse network conditions. Collectively, this thesis presents a comprehensive investigation into protocol performance evolution from TCP to QUIC to HTTP/3, identifies critical challenges in browser protocol management, and introduces an intelligent, adaptive framework to enhance streaming QoE across heterogeneous network conditions.