A semantic framework for event-driven service composition
Metadata[+] Show full item record
Service Oriented Architecture (SOA) has become a popular paradigm for designing distributed systems where loosely coupled services (i.e. computational entities) can be integrated seamlessly to provide complex composite services. Key challenges are discovery of the required services using their formal descriptions and their coherent composition in a timely manner. Most service descriptions are written in XML-based languages that are syntactic, creating linguistic ambiguity during service matchmaking. Furthermore, existing models that implement SOA have mostly middleware-controlled synchronous request/replybased runtime binding of services that incur undesirable service latency. In addition, they impose expensive state monitoring overhead on the middleware. Some newer event-driven models introduce asynchronous publish/subscribe-based event notifications to consumer applications and services. However, they require an event-library that stores definitions of all possible system events, which is impractical in an open and dynamic system. The objective of this study is to efficiently address on-demand consumer requests with minimum service latency and maximum consumer utility. It focuses on semantic eventdriven service composition. For efficient semantic service discovery, the dissertation proposes a novel service learning algorithm called Semantic Taxonomic Clustering (STC). The algorithm utilizes semantic service descriptions to cluster services into functional categories for pruning search space during service discovery and composition. STC utilizes a dynamic bit-encoding algorithm called DL-Encoding that enables linear time bit operationbased semantic matchmaking as compared to expensive reasoner-based semantic matchmaking. The algorithm shows significant improvement in performance and accuracy over some of the important service category algorithms reported in the literature. A novel user-friendly and computationally efficient query model called Desire-based Query Model (DQM) is proposed for formally specifying service queries. STC and DQM serve as the building block for the dual framework that is the core contribution of this dissertation: (i) centralized ALNet (Activity Logic Network) platform and (ii) distributed agentbased SMARTSPACE platform. The former incorporates a middleware controlled service composition algorithm called ALNetComposer while the latter includes the SmartDeal purely distributed composition algorithm. The query response accuracy and performance were evaluated for both the algorithms under simulated event-driven SOA environments. The experimental results show that various environmental parameters, such as domain diversity and scope, size and complexity of the SOA system, and dynamicity of the SOA system, significantly affect accuracy and performance of the proposed model. This dissertation demonstrates that the functionality and scalability of the proposed framework are acceptable for relatively static and domain specific environments as well as large, diverse, and highly dynamic environments. In summary, this dissertation addresses the key design issues and problems in the area of asynchronous and pro-active event-driven service composition.
Table of Contents
Introduction -- Research background -- Semantic service matchmaking & query modeling -- Service organization by learning service category -- ALNet: event-driven platform for service composition -- SMARTSPACE: distributed multi-agent based event-handeling -- Conclusion & future work