Computer Science Department
School of Computer Science, Carnegie Mellon University


Scalable Consistency Management
for Web Database Caches

Charles Garrod, Amit Manjhi, Anastassia Ailamaki,
Phil Gibbons*, Bruce Maggs**, Todd Mowry***,
Christopher Olston****, Anthony Tomasic

July 2006

Keywords: Consistency management, database caching, database scalability services, public/subscribe

We have built a prototype of a scalable dynamic web-content delivery system, which we call S3. Initial experiments with S3 led us to conclude that the key to achieving scalability lay in reducing the workload on back-end databases. S3 utilizes proxy servers to generate dynamic content and cache the results of queries forwarded to the back-end database. This approach introduces the challenge of maintaining cache consistency when the database is updated. In this paper we introduce a fully-distributed consistency management infrastructure that uses a scalable publish / subscribe substrate to propagate update notifications. We use static analysis of the database workload to introduce several design alternatives for how to map database requests to publish / subscribe groups. Finally, we develop a simulation framework and use both simulation and our S3 prototype implementation to evaluate these alternatives empirically to determine which design is best for typical dynamic web workloads.

19 pages

*Intel Research Pittsburgh
**Carnegie Mellon University and Akamai Technologies
***Carnegie Mellon University and Intel Research Pittsburgh
****Carnegie Mellon University and Yahoo Research

Return to: SCS Technical Report Collection
School of Computer Science

This page maintained by