Increasing the Scalability of Dynamic Web Applications

The continued growth of the Web and its increasing role in our daily life has created new technical and social challenges. On the technical side, applications deployed on the Internet suffer from unpredictable load, especially due to events such as breaking news (e.g., Hurricane Katrina) and sudden popularity spikes (e.g., the “Slashdot Effect”). A large number of these Web applications increasingly use a database to generate customized and personalized responses to users’ requests. Because of the widely varying load, currently there is no economical way to provision infrastructure for many of these applications in which the database system is the bottleneck. On the social side, Web applications increasingly collect sensitive data, which must be kept private. In this dissertation we address both these technical and social challenges. We design and implement a Database Scalability Service (DBSS), which can offer scalability to data-intensive applications as a plug-in subscription service with a per-usage charge. A DBSS works by caching applications’ data and answering queries on their behalf. It uses a large shared infrastructure to absorb load spikes that may occur in any individual application. We address two key issues in designing a DBSS: (a) the privacy concerns of applications in allowing the DBSS to cache their data, and (b) the performance concerns due to the high latency applications face in accessing their data in a DBSS setting. Simply encrypting all the data that passes through the DBSS is not a feasible solution to an application’s privacy concerns. On an update, the DBSS must invalidate (at least) data from its cache that have changed . If an application encrypts all the data passing through the DBSS, the DBSS cannot discern any information about what data it is caching. The DBSS then is forced to invalidate large amounts of data from its cache on any update, which leads to poor scalability. In deciding how much data (that passes