Ion Constantinescu

Large scale, type-compatible service composition

Service matchmaking and composition has recently drawn increasing attention in the research community. Most existing algorithms construct chains of services based on exact matches of input/output types. However, this does not work when the available services only cover a part of the range of the input type. We present an algorithm that also allows partial matches and composes them using switches that decide on the required service at runtime based on the actual data type. We report experiments on randomly generated composition problems that show that using partial matches can decrease the failure rate of the integration algorithm using only complete matches by up to 7 times with no increase in the number of directory accesses required. This shows that composition with partial matches is an essential and useful element of Web service composition.

Efficient matchmaking and directory services

It has been widely recognised that matchmaking is an important component for environments populated with heterogeneous services. Several researchers have developed powerful techniques for the matchmaking problem in general. There are also specific representations of service capabilities such as DAML-S, which provide a more specific framework for matchmaking. Most approaches to matchmaking have assumed a sequential search for a service with matching capabilities. This may become intractable when the number of available services gets large. We consider how matchmaking can be developed into service directories that can be searched and maintained efficiently. Our main contribution is to show how matchmaking with DAML-S specifications can be integrated with efficient methods for searching and maintaining balanced directory trees. We also report on experimental results using an implementation based on generalised search trees.

Decentralized Orchestration of CompositeWeb Services

Traditional, centralized orchestration of composite Web services often leads to inefficient routing of messages. To solve this problem, we present a novel scheme to execute composite Web services in a fully decentralized way. We introduce service invocation triggers, a lightweight infrastructure that routes messages directly from the producing service to the consuming one, enabling fully decentralized orchestration. An evaluation confirms that decentralized orchestration can significantly reduce the network traffic when compared with centralized orchestration

Type based service composition

Service matchmaking and composition has recently drawn increasing attention in the research community. Most existing algorithms construct chains of services based on exact matches of input/output types. However, this does not work when the available services only cover a part of the range of the input type. We present an algorithm that also allows partial matches and composes them using switches that decide on the required service at runtime based on the actual data type. We report experiments on randomly generated composition problems that show that using partial matches can decrease the failure rate of the integration algorithm using only complete matches by up to 7 times with no increase in the number of directory accesses required. This shows that composition with partial matches is an essential and useful element of web service composition.

An extensible directory enabling efficient semantic web service integration

In an open environment populated by large numbers of heterogeneous information services, integration is a major challenge. In such a setting, the efficient coupling between directory-based service discovery and service composition engines is crucial. In this paper we present a directory service that offers specific functionality in order to enable efficient service integration. The directory implementation relies on a compact numerical encoding of service parameters and on a multidimensional index structure. It supports isolated service integration sessions providing a consistent view of the directory data. During a session a client may issue multiple queries to the directory and retrieve the results incrementally. In order to optimize the interaction of the directory with different service composition algorithms, the directory supports custom ranking functions that are dynamically installed with the aid of mobile code. The ranking functions are written in Java, but the directory service imposes severe restrictions on the programming model in order to protect itself against malicious or erroneous code (e.g., denial-of-service attacks). With the aid of user-defined ranking functions, application-specific ordering heuristics can be deployed directly. Experiments on randomly generated problems show that they significantly reduce the number of query results that have to be transmitted to the client by up to 5 times.

Directory Services for Incremental Service Integration

In an open environment populated by heterogeneous information services integration will be a major challenge. Even if the problem is similar to planning in some aspects, the number and the difference in specificity of services makes existing techniques not suitable and requires a different approach. Our solution is to incrementally solve integration problems by using an interplay between service discovery and integration alongside with a technique for composing specific partially matching services into more generic constructs. In this paper we present a directory system and a number of mechanisms designed to support incremental integration algorithms with partial matches for large numbers of service descriptions. We also report experiments on randomly generated composition problems that show that using partial matches can decrease the failure rate of the integration algorithm using only complete matches by up to 7 times with no increase in the number of directory accesses required.

Adapting Agent Communication Languages for Semantic Web Service Inter-Communication

The integration of semantic Web and Web services technologies promises to be one of the most promising new areas for development of intelligent Web applications. One challenging area where these technologies meet is in explicit definitions of meaning for the messages exchanged between Web services - in other words, semantic definitions of the meanings of data / commands exchanged in the execution of a Web services based application. While current approaches such as OWL-S tackle these elements in service groundings by mapping processes to function calls with specific arguments, agent communication languages could provide a potentially richer alternative. The work presented here shows how this could be done by mapping the existing agent communication language (FIPA-ACL, FIPA-SL and associated standards developed by the foundation for intelligent physical agents) into OWL based representations which may then be readily used in a Web services environment.

A Directory for Web Service Integration Supporting Custom Query Pruning and Ranking

In an open environment populated by large numbers of heterogeneous information services, integration is a major challenge. In such a setting, the efficient coupling between directory-based service discovery and service composition engines is crucial. In this paper we present a directory service that offers specific functionality in order to enable efficient web service integration. Results matching with a directory query are retrieved incrementally on demand, whenever the service composition engine needs new results. In order to optimize the interaction of the directory with different service composition algorithms, the directory supports custom pruning and ranking functions that are dynamically installed with the aid of mobile code. The pruning and ranking functions are written in Java, but the directory service imposes severe restrictions on the programming model in order to protect itself against malicious or erroneous code. With the aid of user-defined pruning and ranking functions, application-specific ordering heuristics can be directly installed into the directory. Due to its extensibility, the directory can be tailored to the needs of various service integration algorithms. This is crucial, as service composition still needs a lot of research and experimentation in order to develop industrial-strength algorithms. Experiments on randomly generated problems show that special pruning and ranking functions significantly reduce the number of query results that have to be transmitted to the client by up to 5 times.

Optimally distributing interactions between composed semantic web services

When information services are organized to provide some composed functionality, their interactions can be formally represented as workflows. Traditionally, workflows are executed by centralized engines that invoke the necessary services and collect results. If services are clustered (e.g., based on geographic criteria), locally routing intermediary results between services in the same cluster can be more efficient. This paper has several contributions: First, it presents a framework allowing the execution of workflow parts to be mediated by special execution sites. Second, we describe a trigger-based mechanism allowing partial results to be routed between execution sites.Finally, we present an approach for optimally computing the distribution of workflow parts to execution sites accordingly to an integrated cost model for workflow execution. The model is created by merging cost-models provided by individual execution sites trough a Contract Net task brokering protocol. The models consider cost measures for service activation, parameter transfer, and service execution.

A Flexible Directory Query Language for the Efficient Processing of Service Composition Queries

Service composition is a major challenge in an open environment populated by large numbers of heterogeneous services. In such a setting, the efficient interaction of directory-based service discovery with service composition engines is crucial. In this article we present a Java-based directory that offers special functionality enabling effective service composition. In order to optimize the interaction of the directory with different service composition algorithms exploiting application-specific heuristics, the directory supports user-defined selection and ranking functions written in a declarative query language. Inside the directory queries are transformed and compiled to JVM bytecode which is dynamically linked into the directory. The compiled query enables a best-first search of matching directory entries, efficiently pruning the search space.