Reliability prediction in model-driven development

Abstract

Evaluating the implications of an architecture design early in the software development lifecycle is important in order to reduce costs of development. Reliability is an important concern with regard to the correct delivery of software system service. Recently, the UML Profile for Modeling Quality of Service has defined a set of UML extensions to represent dependability concerns (including reliability) and other non-functional requirements in early stages of the software development lifecycle. Our research has shown that these extensions are not comprehensive enough to support reliability analysis for model-driven software engineering, because the description of reliability characteristics in this profile lacks support for certain dynamic aspects that are essential in modeling reliability. In this work, we define a profile for reliability analysis by extending the UML 2.0 specification to support reliability prediction based on scenario specifications. A UML model specified using the profile is translated to a labelled transition system (LTS), which is used for automated reliability prediction and identification of implied scenarios; the results of this analysis are then fed back to the UML model. The result is a comprehensive framework for addressing software reliability modeling, including analysis and evolution of reliability predictions. We exemplify our approach using the Boiler System used in previous work and demonstrate how reliability analysis results can be integrated into UML models. © Springer-Verlag Berlin Heidelberg 2005.