Predicting the Reliability of Component-Based Software Architectures

From the user’s point of view, the reliability of a software component depends on its environment as well as its usage profile. The environment of a component includes the external services invoked by the component and the hardware and software it is deployed on. The usage profile determines which services of the component are needed and describes all possible call sequences in form of a Markov model. The influence of the usage profile and the reliability of external services on the reliability of a componentbased software architecture has been analysed in [38]. There, parametric contracts are used to determine the reliability of a component in its environment. Parametric contracts use so-called service effect specifications which describe the usage of external services by a service provided by the component to create a mapping between the providesand requires interfaces of the same component. We extend the approach described there and consider the reliability of resources like devices (hardware) and execution environments (software). Therefore, we develop a mathematical model to determine the usage period of the resources depending on the usage profile. We compute the reliabilities of the resources on the basis of their usage period. This extends user orientation of software reliability towards system reliability. The consideration of the usage period of a resource requires a mathematical model to determine the execution time of a service. Based on parametric contracts, we develop two approaches to compute the execution time of a service. The first approach makes use of the properties of Markov chains and yields the expected (or average) execution time of a service. The second approach is an extension of parametric performance contracts [37] which describe the execution time of a service in form of a probability density function. We overcome the limits of the approach described there and give a mathematical model to determine the execution time of a loop based on the discrete Fourier transform. Furthermore, we describe how parametric performance contracts can be applied using regular expressions. Furthermore, both computational models are modified to deal with the usage periods of the system resources. The computation of the resource reliability based on the usage period is discussed as well. We use a component-based webserver recently developed in the context of the Palladio project [34] to evaluate some of the predictions made by our model.