A top-down look into the performance of web applications

The Internet ecosystem primarily consists of content publishers who generate web content, end-users who consume the content, and Internet Service Providers who connect them. Since content publishers and Internet Service Providers primarily generate revenue from advertisements and subscriptions, good user experience and high user engagement is of utmost importance. Consequently, measuring and improving the performance of web and the Internet has been an active area of research in the past decade.

In this thesis, we develop techniques to measure the performance of web applications and the underlying network. Notably, we first design scalable and deployable measurement approaches to characterize the performance of networking systems. We analyze the performance aspects of various interconnection strategies adopted by networks, over a large-scale longitudinal measurement study. Additionally, we design a measurement framework that facilitates the deployment of various network measurement techniques with otherwise high network overheads, by leveraging useful application traffic and mitigating the associated network overheads.

In addition to network performance measurements, we also develop approaches to characterize and measure the performance of web applications such as web browsing and video streaming. Specifically, we design a server-side approach to characterize the performance of live video streaming by analyzing incoming traffic at the web servers managed by content providers. We further design an approach to identify the groups of end-users with bad streaming experience in real-time, thereby allowing the stakeholders to take immediate mitigation steps. Next, we design a server-side technique to facilitate the implementation of HTTP/2 server push, allowing content providers to improve the performance of web applications by preemptively sending web content to the end-user without waiting for explicit requests. Finally, we study the usage of standardized HTML resource attributes that control the loading and execution of resources on modern web pages, and their impact on web performance.