Dynamic evolution and reconfiguration of software architectures through aspects

Change is an intrinsic property of software. A software system, during its lifetime, may require several updates, improvements, or new features. If these change requirements are not addressed, the risk of becoming a useless system increases. In fact, this is a challenging issue of safety- and mission-critical software systems, which cannot be stopped to perform maintenance or evolution operations due to their continuous operation. To reduce the aging of these critical systems, they must be provided with mechanisms enabling their dynamic evolution, i.e. the support of changes on their structure and behaviour while they remain in operation. This thesis is concerned with the design of a framework to build architecturebased, dynamically evolvable, software systems. The fact that this framework is a software architecture based approach provides the following advantages: (i) it offers a high-level of abstraction for describing dynamic changes; (ii) it allows varying the level of system description; and (iii) it advantages from the existing support for system modelling, code-generation, and formal analysis provided by architecture description languages. The framework presented in this thesis, called Dynamic PRISMA, is characterized by the combination of two levels of dynamism: Dynamic Reconfiguration, which addresses changes at the configuration level (i.e. the architectural configuration), and Dynamic Type Evolution, which addresses changes at the type-level (i.e. the specification of architectural types and instances). This combination is one of the major contributions of this thesis: thus a system is not only able to reconfigure at runtime the building blocks it is composed of (i.e. architectural types), but also to redefine these building blocks (or introduce new ones) at runtime. Another contribution of the thesis is the identification of the concerns related to dynamic evolution and their integration in the framework through aspects. This improves the separation of concerns and allows us to change reconfiguration specifications, evolution mechanisms, or the business logic independently of each other. A third contribution of this thesis is how this dynamism is supported: reconfiguration through autonomic capabilities, which provides proactivity according to either internal or external stimuli; and type evolution through asynchronous reflection, which enables the modification of a type specification and the transformation of their instances at different rates (i.e. when they are ready for evolution). Specifically, the asynchronous evolution semantics is precisely described by means of graph transformations. This formalism has been chosen because it naturally models both the system architecture and its asynchronous evolution. The work presented in this thesis is illustrated through a case study from the robotics domain; an area which could potentially benefit from the results of this thesis. KEYWORDS software architectures, dynamic evolution, dynamic reconfiguration, dynamic type evolution, dynamic instance evolution, asynchronous evolution, runtime evolution, self-managed systems, autonomic computing, computational reflection, aspect-oriented software development