We are investigating approaches to rapidly and efficiently composing new web services. There are two main challenges to achieve this objective. First, a technique is needed for efficient integration of multiple web sources. Second, as the system scales up and more and more web service objects (i.e., sources and their corresponding objects) are introduced, we need a scalable architecture to evenly distribute the load on our web-service. We are investigating these challenges through the following two subprojects using .NET:

1. An approach that rapidly integrates different web-accessible information sources. This approach supports semi-structured data by using wrappers that extract information from web sources and provide a query interface to the data.  The resulting information agents can perform various integration, monitoring and notification tasks on the Internet (or an Intranet setting), and, in turn, constitute new web services. This project builds on existing .NET capabilities such as the Simple Object Access Protocol (SOAP), the Web Services Description Language (WSDL), and the Universal Description, Discovery, Integration (UDDI) specification.

2. In order to dramatically scale up the number of supported web services, we are developing a middleware that manages placement of web service objects on a cluster of nodes, providing its target application the illusion that objects are managed in a centralized manner. This middleware monitors and adjusts the placement of object in order to respond to an application's changing workload characteristics. It utilizes a Dependable Web (DeW) framework to resolve dangling proxy references when a web service migrates. (A natural extension is to incorporate GXA's WS-Referral and WS-Routing which we propose to investigate in 2003.) It employs the remoting capabilities of .NET, its web applications, and operating system services to monitor resources and migrate objects. This middleware minimizes the overhead of communication by compressing messages when appropriate, i.e., when the overhead of compression does not outweigh its benefits. We are implementing and evaluating a prototype of this middleware using C#.

We plan to instrument the web services produced by the first effort to gather user trace data. We use these traces to investigate the techniques proposed by the second effort using large-scale network simulation models.