CRII: NeTS: Making Sense of Mobile Web Page Performance

Mobile traffic is soon poised to exceed non-mobile traffic across the world. Already, for many users, mobile phones are the only 'computer' they own. As a result, mobile Web pages are fast becoming the primary portal for content. However, loading a Web page on the mobile phone is exceedingly slow. There have been several optimizations designed to speed up Web pages, but there is no silver bullet: an optimization that speeds up one page can, and often does, slow down another page. One of the goals of this project is to shed light on why Web pages load slower on the phone as compared to a desktop. The project will use this bottleneck analysis to design algorithms, techniques, and tools, to accurately estimate if any given optimization will help or further slow down the page load time. The end goal of the project is a comprehensive analysis framework called ProfX that will help Web developers, network providers, browser vendors, and researchers design and choose the right set of optimizations to significantly speed up the mobile Web. The problem is that the mobile Web ecosystem is fairly complex. This makes understanding the page load bottlenecks and analyzing optimizations challenging. Second, the time it takes to load a page depends on various factors. Isolating the effect of an optimization under such variability is non-trivial. To address these challenges, the project proposes two key ideas: (1) Piecemeal analysis: Web page loading is not an atomic activity. The page load process is made up of several smaller activities that form a complex dependency structure. The project proposes to decompose the Web page process and analyze each sub activity to identify the bottlenecks. Further, the effect of an optimization will be studied on each of these sub activities. This piecemeal analysis addresses the complexity challenge in the Web ecosystem, and (2) Isolation: The project proposes to develop techniques and experimental tools to control the variability of the page load process. The goal is to isolate the effect of a given optimization by forcing the effect of other unrelated parameters to be a near constant. The techniques developed in this project will be more broadly applicable to studying other mobile applications such as video.