Toward Automatic State Management for Dynamic Web Services

A key challenge in the development of the Internet is to simplify construction of scalable widearea services. One approach to scaling wide-area services is to deploy generic computing power and storage in the network, and use it to absorb service load through dynamic resource recruitment, active caching, or dynamic service replication. Each of these approaches introduces distributed state and an accompanying burden on the programmer to manage that state. This paper develops an approach to automatic state management for replicated services, a key step toward the goal of automatically converting unscalable service implementations into scalable ones. We demonstrate a prototype implementation of automatic state management, called Ivory. Ivory transforms the bytecodes of a Java-based service to trap updates to its data structures and propagate modi ed objects among to peer replicas. We demonstrate our approach in the context of a service caching framework that replicates service code and data on demand, and present measurements of an example Web portal application that shows the overhead and scalability bene ts of service replication using Ivory.