Specification and Analysis of Availability for Software-Intensive Systems

For many technical systems, their availability is an important characteristic. Examples for systems, where availability is especially relevant, are railway control systems, telecommunication systems, and systems supporting business processes. Outages of such systems often have immediate economic consequences for their users and operators. Due to the economic impact, manufacturers and service providers need to guarantee a certain level of availability and these guarantees become part of the system requirements. To avoid high costs due to changes in late development phases, availability requirements need to be specified and verified as early as possible. To enable early predictions of the system availability, a range of modeling techniques have been proposed by academia and are in use in practice. However, current modeling techniques have two drawbacks: First, they describe the system on a high level of abstraction, usually in terms of their architecture. This makes it hard to formulate meaningful, system-specific availability requirements that relate to the functional requirements of the system. Second, the current techniques are not embedded into a comprehensive engineering method, which defines relationships between different models and provides an engineering process. In this thesis, we provide evidence for the stated problems and propose a solution. First, we report on a qualitative interview study we conducted with 15 industrial availability experts from different domains. With this study, we assess the relevance of the topic availability in the industry, determine availability related activities and methods, and identify several problems connected to the specification and verification of availability requirements. Second, as a solution to the stated problems, we extend an existing artifact model for software-intensive systems by additional artifacts supporting the concise specification and analysis of system-specific availability properties. The first additional artifact is an availability requirements specification. It uses concepts from the second artifact, the availability specification, which captures system-specific definitions of failure and availability metrics. The third artifact is an extended logical architecture that includes the system’s behavior in case of faults. The last artifact is an environment specification, which contains the structure and behavior of the system’s environment. For each additional artifact, we suggest suitable models and description techniques to capture the necessary information. Third, we provide a modeling method that supports the systematic application of our artifact model. The modeling method consists of basic modeling building blocks, a process for instantiating the artifact model, step-by-step guides for systematically creating individual models, and modeling patterns providing a basic structure for some of the model types. Finally, we evaluate our artifact method and our modeling method in an industrial case study. In the case study, we assess the adequacy and the flexibility of our modeling approach. We further report on prototypical tool support and evaluate the feasibility of an availability analysis based on the created models. To perform the case study, we model a section of an industrial train control system, departing from the original requirements. We then extend this initial model by instantiating our availability artifact model. As a last step, we perform several types of availability analyses.