Evolutionary Environmental Modelling in Self-Managing Software Systems

The promise of robust software that can self-manage significant aspects of its operation, including the ability to self-configure, self-heal, self-optimise and self-protect through having the requisite functionality to respond and adapt to changes in its operational environment is both seductive and compelling. There are a growing number of examples of partial implementations appearing in the literature and continued development across a number of areas can be expected in the future.One of the less travelled areas of research concerns the problem of developing an accurate and current model of the environment in which such adaptive systems will operate. It would seem a compelling argument that holding a current model of both the environment and the current capability of the system allowing the system to "know itself" are desirable additions to any adaptive system. As such they have a view of the complex space within which they can adapt and that without these properties the system could only be considered as purely reactive.Here, the use of Learning Classifier Systems and geneticalgorithms to provide the modelling element required of effective adaptive software systems is presented and evaluated. The work uses the virtual world platform of "Second Life" to represent anappropriate experimental environment. One outcome of this work is the restatement of some classical cybernetic principles to reflect the need for constant evolution.