Towards Probabilistic Models to Predict Availability, Accessibility and Successability of Web Services

Web Services are gaining increasing attention as programming components and so is their quality. The external qualities of Web Services (i.e., qualities that are perceived by their users) such as the OASIS sub-quality factors Availability, Accessibility, and Successability can only be measured at late stages after the deployment and the provisioning of the Web Service. This may necessitate expensive rework if the targeted levels of qualities are not satisfactorily met. A reliable prediction of the values of the external qualities at early phases during development may totally remove the need for reworking and hence save valuable resources. In this paper, we describe an approach for building and empirically evaluating probabilistic prediction models for the Web Services external sub-quality factors Availability, Accessibility, and Successability based on internal static and dynamic quality measures (e.g., Cyclomatic Complexity and Distinct Method Invocations). A methodology was established that involves the collection of a set of predefined quality measures and then performing regression analysis to identify any correlation between them and the above mentioned external qualities. For this purpose, a framework for data collection and evaluation was designed, implemented and tested. The results of the preliminary evaluation of the framework showed that it is feasible to collect all the data points necessary for the regression analysis and model building activities. We are currently working towards adding about 18 more Web Services to our testbed in order to carry out a wider controlled experiment and then to build possibly accurate probabilistic prediction models for Availability, Accessibility, and Successability.