Minimizing feature model inconsistencies in Software Product Lines

Abstract

Software Product Line (SPL) is a software engineering methodology to create and manage a family of similar software products by using reconfigurable feature models. In a large-scale SPL, selection of the relevant set of features for configuring a given product is a key challenge, each software unit is configured by a feature set and combining features from each unit can generate inconsistencies which are solved by manual deliberation between system designers, leading to possible loss of valuable business resources. In this paper, we employ Genetic Algorithms (GA) to minimize three primary feature model inconsistencies, i.e., mandatory, inclusive and exclusive/alternative, with a scattered cross-over function and 1% mutation rate. Using real-world feature models from a local smart phone SPL, we optimize a small-scale feature model (containing 100 features) and two large-scale ones (containing 500 and 1000 features) and show that GA can produce up to 95–97% consistent (conflict-free) feature models in drastically reduced times as compared to manual conflict resolution techniques. We also show that a scattered cross over function produces better results than single-point or multi-point functions. While slightly increasing the mutation rate improves the overall optimality of the solution.