Project R-7710

Web Performance Automation for The People (Research)

Modern websites are often slow to load and interact with. This is so annoying to many users, that every research into the topic finds that slow load times directly correlate with loss of revenue. For high profile websites, even a slowdown of 100ms can lead to hundreds of thousands of dollars of missed profit. Even though there are already many optimization solutions available (both industrial and academic) to help with this problem, many websites do not use them or only a subset of them and even then they often fail to get the best possible performance. This is mainly because the optimizations are difficult to implement and evaluate and because they are often very context dependent and should thus be fine-tuned for different situations. A last aspect is that users have no way to indicate their preferences when it comes to how websites should load and which aspects they find most important, so site owners have to serve the same version to every visitor. With the recent arrival of the HTTP/2 protocol and its contradictory results, it is now more clear than ever that Web Performance Optimization (WPO) is very difficult to implement and maintain. We propose to move towards a more automated approach for WPO, where an intermediate server applies existing complex optimizations itself with minimal effort from the website developer. This server then also takes into account user preferences and settings to deliver highly specialized versions of the website for each individual context.

01 January 2017 - 31 December 2020