Date of Award

Spring 2025

Project Type

Dissertation

Program or Major

Computer Science

Degree Name

Doctor of Philosophy

First Advisor

Radim Bartos

Second Advisor

Maanak Gupta

Third Advisor

Ying Li

Abstract

The internet is an indispensable part of modern life, with billions of users accessing countless web pages daily. While loading speed is crucial, the overall Quality of Experience (QoE) significantly influences user satisfaction and engagement with websites. Incomplete or delayed page loading can lead to frustration, reduced site visits, and financial losses. In our preliminary study, we examined how network protocols like HTTP/2 and HTTP/3 impact both Quality of Service (QoS) and QoE by assessing throughput and six Lighthouse performance metrics. Our findings highlight the relationship between QoS and QoE, emphasizing the need for efficient resource scheduling and prioritization to enhance user experience.

Network protocols like HTTP/3 and QUIC are crucial for delivering web resources and thus influence QoE. The Extensible Prioritization Scheme (EPS) allows clients and servers to indicate resource priority hints, enabling data to be delivered either incrementally or non-incrementally. However, their growing complexity poses challenges in resource (documents, stylesheets, fonts, images, and etc) scheduling and prioritization.

First, we address the incremental delivery (IP) strategy, where the main challenge is appropriately allocating bandwidth among resources. We propose a weighted incremental scheduling mechanism to effectively distribute bandwidth to various web resources.

Next, we examine non-incremental delivery (NIP), where each resource is delivered in its entirety. This can cause Head-of-Line (HoL) blocking, where larger resources delay smaller, more important ones. To solve this, we propose an urgency-based non-incremental delivery mechanism that controls the order of resource delivery based on their urgency complemented with the resource-type-aware mapping.

To overcome the limitations of incremental and non-incremental resource delivery, we design a combined approach called the MIX delivery mechanism, integrating both incremental and non-incremental methods. The MIX delivery mechanism balances the two strategies by dynamically switching between IP and NIP while respecting their urgency levels for delivering resources.

We evaluated these mechanisms using a network testbed that simulates real-world conditions. Utilizing a diverse set of popular websites and assessing performance with Google's Lighthouse tool, our experimental results show that the IP and NIP scheduling mechanisms perform better than the default sequential (SEQ) scheduling mechanism. Moreover, the MIX scheduling mechanism outperforms all others and further enhances QoE.

Download

Share

COinS