OntoArch Reliability-Aware Software Architecture Design and Experience

Background And Motivations

Traditionally, reliability analysis is performed after system’s implementation, when corrections are difficult and modifications are time consuming and expensive. Currently the interest of reliability evaluation has turned to quality evaluation at the software architecture level. Several proposals have been made to predict reliability at the architecture level. In (Immonen & Niemelä, 2007) we compare six most promising reliability evaluation methods, i.e. the methods from Cortellessa et al. (Cortellessa, Singh, & Cukic, 2002), Rodrigues et al. (Oreizy & Taylor et al., 1999), Yacoub et al. (Yacoub, Cukic, & Ammar, 1999), Reussneer et al. (Reussner, Schmidt, & Poernomo, 2003), and Wang et al. (Wang, Pan, & Chen, 1999) from the viewpoint of software architecture. In summary, all the surveyed methods require additional work, especially in the development of the analysis model and application of mathematical algorithms. Although many of these methods have been created for years ago, only few evidences of their maturity are available; i) the methods have been experimented only in laboratory settings by the authors; ii) no comparison of the predicted and measured reliability values exists; and iii) no cost estimation of using the methods could be found. The above mentioned methods are targeted to system architects, software architects, software integrators or service assemblers. It is also common to these methods that traceability of reliability requirements to software architecture is missing. Moreover, most of these methods lack tool support for reliability evaluation. Thus, we conclude the main findings as a motivation for our approach; methods that apply UML, like (Cortellessa et al., 2002; Rodrigues et al., 2005; Yacoub et al., 1999), in reliability analysis models have several benefits; (i) no additional learning effort because architects are familiar with UML, (ii) no extra modeling cost because no specific analysis model is required, and (iii) open source and commercial tools are available. Therefore, our approach heavily relies on the use of common and widely accepted modeling constructs in architecture models that are extended with required and measured reliability values (Ovaska & Evesti et al., 2010).