Evolutionary environmental modelling in self-managing software systems

Over recent years, the increasing richness and sophistication of modem software systems has challenged conventional design-time software modelling analysis and has led to a number of studies exploring non-conventional approaches particularly those inspired by nature. The natural world routinely produces organisms that can not only survive but also flourish in changing environments as a consequence of their ability to adapt and therefore improve their fitness in relation to the external environments in which they exist. Following this biologically inspired systems' design approach, this study aims to test the hypothesis - can evolutionary techniques for runtime modelling of a given system's environment be more effective than traditional approaches, which are increasingly difficult to specify and model at design-time? This work specifically focuses on investigating the requirements for software environment modelling at runtime via a proposed systemic integration of Learning Classifier Systems and Genetic Algorithms with the well-known managerial cybernetics Viable Systems Model. The main novel contribution of this thesis is that it provides an evaluation of an approach by which software can create and crucially, maintain a current model of the environment, allowing the system to react more effectively to changes in that environment, thereby improving robustness and performance of the system. Detailed novel contributions include an evaluation of a variety of environmental modelling approaches to improving system robustness, the use of Learning Classifier Systems and genetic algorithms to provide the modelling element required of effective adaptive software systems. It also provides a conceptual framework of an Environmental Modelling, Monitoring and Adaptive system (EMMA) to manage the various elements required to achieve an effective environmental control system. The key result of this research has been to demonstrate the value of the guiding principles provided by the field of cybernetics and the potential of Beer's 2 cybernetically based Viable System Model in providing a learning framework, and subsequently a roadmap, to developing self-managing autonomic systems. The work is presented using a virtual world platform called "Second Life". This platform was used for experimental design and testing of results