A Petri-net-based Approach To Modeling And Reliability Determination Of Ontology-b Ased Semantic Web Service Compositions

Today, the Internet provides us access to a large number of Web services, most ofwhich are for specific functions. However the performance of a single service is verylimited. If a method can be found to enable us to describe web services in a waycomputer can understand, we will be able to get value-added services by combiningsingle-function services, and this is what the semantic web researches.OWL-S is one of the most effective languages of semantic web service. It providesan ontology-based frame and rules following which the description of web service cangain some semantic information, so that the computer can understand our descriptionin some way. OWL-S is convenient to use and has strong describing abilities. However,as the current theoretical research in semantic language focuses on its formalizationand property verification, the non-functional properties and quantification analysis isrelatively weak. In order to enable users to determine whether a service can meet theirrequirements for non-functional characteristics, and to choose the most beneficial oneto their system, we need to establish a method to analyze the service quality ofcomposite services in OWL-S.In this paper, we establish a model of the atomic service and the structure serviceinvolved in OWL-S, based on non-Markovian stochastic Petri net (NMSPN).Using thismodel, we can calculate the PNCP (process normal completion probability) of a singleservice by probability-based method, and obtain the reliability of the whole system byfurther calculations. This method takes probabilistic parameters of the serviceinvocations and the SOAP messages as model inputs.In order to validate the feasibility and accuracy of this method, we analyze aninstance thoroughly. First we create a Petri model of the system and obtain a variety ofruntime data, therefore we can calculate the reliability of the system, and finally weconduct a confidence interval analysis of the results. In the last part of the paper, asensitivity analysis is also performed to determine the impact of different modelparameters on the system reliability and help to identify the reliability bottlenecks. Sowe can know which service is of crucial influence to the overall systems, and thenfocus on the optimization of this service to enhance the reliability.