Task Specification and Reasoning in Dynamically Altered Contexts

Software systems are prone to evolution in order to be kept operational and meet new requirements. However, for large systems such evolution activities cannot occur in a vacuum. Instead, specific action plans must be devised so that evolution goals can be achieved within an acceptable level of deviation or, risk. In this paper we present an approach that allows for the identification of plans in the form of actions that satisfy a goal model when the environment is constantly changing. The approach is based on sequences of mutations of an initial solution, using a local search algorithm. Experimental results indicate that even for medium size models, the approach outperforms in execution time the weighted Max-Sat algorithms, while it is able to achieve an almost optimal solution. The approach is demonstrated on an example scenario of re-configuring a dynamically provisioned system.