Improving HTTP/3 Quality of Experience with Incremental EPS

TL;DR

The paper addresses QoE for web content delivery over HTTP/3 by introducing weighted incremental scheduling based on the Extensible Prioritization Scheme (EPS) with a tunable α. It implements the mechanism in an aioquic HTTP/3/QUIC stack and evaluates performance on eight sites using Lighthouse metrics, revealing overall improvements in FCP, LCP, SI, and CLS, particularly at higher α, while noting site-specific tradeoffs such as TTI and TBT increases. The contributions include a concrete EPS-based weighting formula $w'_i = α/(u_i + r_u) + (1-α)/n$ and normalized $w_i = w'_i / \sum_{j=1}^{n} w'_j$, plus an α-driven balancing framework and practical mapping to browser priority signals. The findings demonstrate the practical potential of EPS for QoE gains in HTTP/3, while highlighting that page structure strongly influences outcomes and that future work should explore dynamic priority updates and hybrid loading strategies to further improve robustness.

Abstract

With the introduction of QUIC, a modern transport-layer network protocol, HTTP/3 leverages its benefits to enhance web content delivery. This paper proposes a mechanism based on the recently standardized Extensible Prioritization Scheme (EPS) for weighted incremental web content delivery. The mechanism augments the sequential scheduling to provide incremental and weighted incremental resource delivery. An existing HTTP/3 implementation was extended with the proposed mechanism and tested with the content of eight popular websites. The results of our experimental analysis show that weighted incremental prioritization improves Quality of Experience (QoE) as measured by Lighthouse, a standard QoE test tool. While overall improvements were generally achieved, we also observed a few cases where the performance degraded slightly, highlighting that the QoE is sensitive to factors such as web page structure.

Figures (5)

• Figure 1: Experimental Setup
• Figure 2: Number and Size of Downloaded Websites
• Figure 3: Percentage Metric Improvements over Sequential Delivery for Varying $\alpha$ Values
• Figure 4: LCP Improvement Trends ($\alpha = 1$)
• Figure 5: Comparison of web performance metrics across different websites.