Proteus: A System for Execution of Dynamically Composed Web Services Using GXA

A. Executive Summary The goal of this project is to utilize the GXA framework (i.e., WS-Inspection, WS-Routing, and WS-Referral) to develop Proteus, a system to: a) dynamically compose plans that integrate web services, b) execute a plan as efficiently as possible in the presence of failures and web service migrations, and c) monitor and show the status of different components at runtime. As a motivating example, consider the problem of identifying a building in an image. This can be done by combining web services for imagery (i.e.,TerraService) with services for the property tax sites and online phone books (see the description of our previous work, Section B). One could write a program to integrate information from the appropriate web services to solve this query for a given area, but the challenge is that there are approximately a thousand property tax sites and hundreds of telephone books for the US and each of them has different levels of coverage. For example, in New York State there is one tax service, but in California there are dozens. A better alternative is for a system such as Proteus to dynamically identify, compose, and execute the appropriate web services to process a query. First, Proteus would identify the relevant web services. In our example, relevant web services would include Microsoft’s TerraService for imagery, the property tax and telephone book services for the given area, and a geocoding service to convert street addresses into lat/long coordinates. Second, it identifies the most efficient plan and executes it to produce a timely response. Third, it monitors and controls the execution of a plan in support of physical-locationindependence, which means the plan will execute as long as a copy of the referenced web services is available. This criterion is important because it frees the end user (and programmers) by requiring the system to resolve the location of a web service in the presence of both a) web service migrations to balance load, and b) node failures that render a copy of a web service unavailable. In order to monitor the execution of a plan, Proteus will provide visualization tools that query the run-time components for their status. This project will utilize our expertise with Microsoft .NET framework and C# programming language. In addition, it utilizes our research results from a 2002 Microsoft .NET research gift (see Section B, Microsoft’s technical contact: Dan Fay, see URL: http://dblab.usc.edu/WebServices for details) that investigated key topics such as turning semi-structured sources into web services, combining web services [Thakkar et al., 2002], and techniques to efficiently execute those services using compression [Cai et al. 2002, Ghandeharizadeh et al. 2002b], streaming, and a Dependable Web service (DeW) framework [Ghandeharizadeh et al. 2002c]. We propose to build upon this foundation to investigate the following challenges: 1. Employ GXA’s WS-Inspection to locate an appropriate set of web services to enable the system to integrate services published by others or even by the user of the system. 2. To extend the WS-Inspection representation language and use it to describe the capabilities of a given web service. These descriptions will provide sufficient detail on the sources to support the automatic composition in Step 3. 3. To apply and extend our previous work on dynamic information integration [Arens et al. 1996; Knoblock et al. 2001] to the problem of automatically composing the available web services in response to specific requests. The automatically generated composition plan will be expressed as a WS-Routing plan, whose execution will be optimized in Steps 4 and 5. 4. To utilize techniques for both streaming data between web services and compressing large XML documents in order to effic iently execute web services across large data sources. 5. To employ the Dependable Web (DeW) framework to support physical-location-independence. This framework utilizes WS-Referral specification to dynamically re-route requests.