View-Driven Optimization of Database-Backed Web Applications

This paper describes HYPERLOOP, a system for optimizing databasebacked web applications (DBWAs). Current approaches in optimizing DBWAs focus on partitioning the application among the browser, application server, and the database, and rely on each component to optimize their portion individually without developer intervention. We argue that this approach misses the goal of DBWAs in optimizing for end-user experience, and fails to leverage domain-specific knowledge that DBWA developers have. For instance, a news website might prioritize loading of the news headlines, even at the expense of slowing down loading of other visual elements on the page. HYPERLOOP illustrates the idea of viewdriven optimization by allowing developers to specify priorities for each of the elements on the webpage, and uses such information to drive optimization of the entire webpage. HYPERLOOP currently focus on optimizing for render time of webpage components, and our preliminary results show that this view-driven approach can substantially improve DBWA performance by leveraging developer provided application knowledge.