A Methodology and Framework for Model-Driven Dependability Analysis of Critical Embedded Systems and Directions Towards Systems of Systems

In different domains, engineers have long used models to assess the feasibility of system designs; over other evaluation techniques modeling has the key advantage of not exercising a real instance of the system, which may be costly, dangerous, or simply unfeasible (e.g., if the system is still under design). In the development of critical systems, modeling is most often employed as a fault forecasting technique, since it can be used to estimate the degree to which a given design provides the required dependability attributes, i.e., to perform quantitative dependability analysis. More in general, models are employed in the evaluation of the Quality of Service (QoS) provided by the system, under the form of dependability, performance, or performability metrics. From an industrial perspective, modeling is also a valuable tool in the Verification & Validation (V&V) process, either as a support to the process itself (e.g., FTA), or as a means to verify specific quantitative or qualitative requirements. Modern computing systems have become very different from what they used to be in the past: their scale is growing, they are becoming massively distributed, interconnected, and evolving. Moreover, a shift towards the use of off-the-shelf components is becoming evident in several domains. Such increase in complexity makes model-based assessment a difficult and time-consuming task. In the last years, the development of system has increasingly adopted the Component-Based Development (CBD) and Model-Driven Engineering (MDE) philosophies as a way to reduce the complexity in system design and evaluation. CBD refers to the established practice of building a system out of reusable “black-box” components, while MDE refers to the systematic use of models as primary artefacts throughout the engineering lifecycle. Engineering languages like UML, BPEL, AADL, etc., allow not only a reasonable unambiguous specification of designs, but also serve as the input for subsequent development steps like code generation, formal verification, and testing. One of the core technologies supporting model-driven engineering is model transformation. Transformations can be used to refine models, apply design patterns, and project design models to various mathematical analysis domains in a precise and automated way. In recent years, model-driven engineering approaches have been also extensively used for the analysis of the extra-functional properties of the systems. To this purpose, language extensions were introduced and utilized to capture the required extra-functional concerns. Despite several approaches propose model transformations for dependability analysis, still there is not a standard approach for performing dependability analysis in a MDE environment. Indeed, when targeting critical embedded systems, the lack of support for dependability attributes, and extra-functional attributes in general, is one of the most recognized weaknesses of UML-based languages. Also, most of the approaches have been defined as extensions to a "general" system development process, often leaving the actual process unspecified. Similarly, supporting tools are typically detached from the design environment, and assume to receive as input a model satisfying certain constraints. While in principle such approach allows not to be bound to specific development methodologies, in practice it introduces a gap between the design of the functional system model, its enrichment with dependability information, and the subsequent analysis. Finally, the specification of properties our of components' context, which typically holds for functional properties, is much less understood for non-functional properties. The work in this thesis elaborates on the combined application of the CBD and MDE philosophies and technologies, with the aim to automate dependability analysis of modern computing systems. A considerable part of the work described in this thesis has been carried out in the context of the ARTEMIS-JU “CHESS” project, which aimed at defining, developing and assessing a methodology for the component-based design and development of embedded systems, using model-driven engineering techniques. The work in this thesis defines and realizes an extension to the CHESS framework for the automated evaluation of quantitative dependability properties. The extension constitutes of: i) a set of UML language extensions, collectively referred to as DEP-UML, for modeling dependability properties relevant for quantitative analysis; ii) a set of model-transformation rules for the automated generation of Stochastic Petri Nets (SPNs) models from system designs enriched with DEP-UML; and iii) a model-transformation tool, realized as a plugin for the Eclipse platform, concretely implementing the approach. After introducing the approach, we detail its application with two case studies. While for embedded systems it is often possible, or even mandatory, to follow and control the whole design and development process, the same does not hold for other classes of systems and infrastructures. In particular, large-scale complex systems don’t fit well in the paradigm proposed by the CHESS project, and alternative approaches are therefore needed. Following this observation, we then elaborate on a workflow for applying MDE approaches to support the modeling of large-scale complex systems. The workflow is based on a particular modeling technique, and a supporting domain-specific language, TMDL, which is defined in this thesis. After introducing a motivating example, the thesis details the workflow, introduces the TMDL language, describes a prototype realization of the approach, and describes the application of the approach to two examples. We then conclude with a discussion and a future view on how the contribution of this thesis can be extended to a comprehensive approach for dependability and performability evaluation in a "System of Systems" context. More in detail, this dissertation is organized as follows. Chapter 1 introduces the context of the work, describing the main concepts related to dependability, and dependability evaluation, with a focus on model-based assessment. The foundation of CBD and MDE approaches, the role of the UML language, and main related work are instead discussed in Chapter 2. Chapter 3 describes the CHESS project, and introduces the language extensions that have been defined to support dependability analysis. Moreover, the chapter details the entire process that drove us to such extensions, including the elicitation of language requirements and the evaluation of existing languages in the literature. The model-transformation algorithms for the generation of Stochastic Petri Nets are described in Chapter 4, while the adopted architecture for the concrete realization of the analysis plugin is described in Chapter 5. Chapter 6 describes the application of our approach to two case studies: of a multimedia processing workstation and a fire detection system. The need for a complementary approach for the evaluation of large-scale complex system is discussed in Chapter 7, with the aid of a motivating example of a distributed multimedia application. Chapter 8 describes our approach for the automated assembly of large dependability models through model-transformation. The thesis then concludes with an outlook on the relevance of the work presented in this thesis towards a System of Systems approach to the evaluation of large-scale complex systems.