Content area
Nowadays, database-backed software is widely used for online shopping, social networking, and many others, where increasingly huge amount of user data is managed and processed. They are often built three-stack architecture: (1) a web interface developed by markup language like HTML (2) application logic developed by traditional object-oriented languages, and (3) a database management system (DBMS) that maintains persistent data. Object Relational Mapping (ORM) frameworks have become increasingly popular to help developers to construct web applications. These ORM frameworks allow developers to program such database-backed web applications in a DBMS oblivious way, as the frameworks expose APIs for developers to operate persistent data stored in the DBMS as if they are regular heap objects, with regular-looking method calls transparently translated to SQL queries by frameworks when executed.
Under this architecture, there are common understanding gaps between web designer and application developers and database engine. As a result, performance and correctness problems are widely existing.
To tackle performance problems, we did one of the first studies to understand why real-world web applications are slow. Guided by that study, we used cross-stack analysis to synthesize efficient web pages, to automatically detect inefficiency in data processing code, and to help database optimization using application knowledge. Our tools have found thousands of performance issues. To tackle correctness problems in web applications, on the one hand, we investigated how data constraints could be inconsistent across web interface, application, and database, and how to solve these problems. On the other hand, we investigated how frequently data schemas evolve and developed EvolutionSaver, which is a static code analysis and transformation tool that automates schema-related code refactoring and consistency checking. The work in the dissertation has already made impact: the empirical study of performance problems has raised attention in open source community. The tools developed have detected and fixed thousands of performance and correctness issues in popular web applications.