SpiderNet: A Quality -Aware Service Composition Middleware

135 p.Thesis (Ph.D.)--University of Illinois at Urbana-Champaign, 2004.The result of our research investigation is SpiderNet, a quality-aware service composition middleware framework. SpiderNet adopts a hybrid system architecture that consists of a core SpiderNet and multiple access SpiderNet subsystems. The major contributions of SpiderNet are as follows. First, we introduce a novel service overlay network model for constructing the core SpiderNet. The service overlay network connects previously dispersed service components via application-level connections to achieve high-quality and failure-resilient service composition. Second, we present two different designs for the core SpiderNet: utility SpiderNet and P2P SpiderNet, which are designed for managed enterprise systems and self-organized peer-to-peer systems, respectively. The utility SpiderNet employs global-state-based centralized approach for initial service composition and reactive failure recovery for maintaining the QoS of composed services during service runtime. In contrast, the P2P SpiderNet provides local-state-based fully distributed solution for initial service composition and proactive failure recovery for runtime service maintenance. Third, we present the access SpiderNet for ubiquitous service delivery. The access SpiderNet provides both service adaptation and service distribution to address the challenges of user mobility, device heterogeneity, and resource constraints in ubiquitous computing environments. We evaluate and validate the SpiderNet system both analytically and experimentally. We have conducted extensive experiments using both large-scale simulations and PlanetLab-based prototype implementation. We also implement a set of distributed multimedia services on top of SpiderNet as proof-of-concept application examples. Our experimental results demonstrate the implementation feasibility and performance efficiency of the SpiderNet system.U of I OnlyRestricted to the U of I community idenfinitely during batch ingest of legacy ETD