Content area
This dissertation presents advanced proxy caching techniques for database-backed web sites. While web caching proxy servers are essential for improving web performance and scalability, their current implementations, which only cache static web objects, are ineffective for database-backed web sites. Since a substantial portion of web traffic consists of queries to database-backed web sites, it is important to enhance the proxy servers so that they can provide caching for these web-based database applications.
My thesis states that proxy caching techniques beyond exact-match URL matching are feasible and useful for web-based database applications. The key insight in this work is that, in order to share the workload of database-backed web sites, it is necessary to add a query processing capability to web proxies and application servers. By distributing the database capability to the large number of edge servers on the Internet, we can potentially scale up a database-backed web site to handle arbitrarily heavy workloads.
The three questions addressed in this dissertation are (i) how we add a query processing capability to web proxies or application servers, (ii) what query processing capability we add, and (iii) how these enhanced proxy servers perform. Correspondingly, we make the following three contributions. Firstly, we demonstrate that a query processing capability can be added to a proxy by having the web server send the proxy a piece of Java code dynamically, or by collocating a query processing engine with the proxy statically. Secondly, we show that the query processing capability added to the proxy can range from query containment checking and simple query processing that take advantage of the characteristics of web queries in lightweight approaches, to full SQL processing in industrial-strength solutions. Finally, we find that even the overhead of adding a full-fledged DBMS to an application server and that of propagating updates from the backend to the cache was insignificant under heavy loads. Consequently, deploying multiple proxies or application servers each with a database cache improves the performance and scalability of database-backed web sites.