Optimized Feature Selection and Variability Management of Feature Models in Software Product Line

Abstract

A Software Product Line (SPL) is used for the development of a family of products that utilize the reusability of existing resources from core assets. Core assets consists on common and variable features. Feature model is extensively used to manage the common and variable features of a family of products. Common features are easy to manage due to part of every product without any change; however, variable features are hard to manage due to environmental selection in product, complex constraints and relationships in feature model. Organizations use the SPL for the development of products with low costs and time to market due to the reusability of common and variable resources. To adopt SPL, organizations require information such as cost, scope, complexity, number of features, the total number of products and the combination of features for each product to start the application development. All possible combinations of features make it easy to find the cost of complete SPL and individual products. Furthermore, the combination of features for each product enables the easy development of products. However, selection of features according to market segments such as cost, performance is challenging task in large and complex constrained feature model. In the large-scale feature model, manual selection of features is the time-consuming and error-prone task. Therefore, optimization enables to find the optimum combinations for product development according to end-user perspective. Due to the high probability of relationship constraint violations, obtaining optimum feature combinations from large feature models is a challenging task. In this dissertation, first, we have proposed Binary Pattern for Nested Cardinality Constraints (BPNCC) approach, which is the simple and effective approach to calculate the exact number of products with complex relationships between application’s feature models. Furthermore, BPNCC approach identifies the feasible features combinations of each product by tracing the constraint relationship from top-to-bottom. BPNCC approach is an open source and tool-independent approach that does not hide the internal information of selected and non-selected features. The proposed method is validated by implementing it on small and large feature models with ‘‘n’’ number of constraints, and it is found that the total number of products and all features combinations in each product without any constraint violation. Furthermore, in order to obtain optimum solutions, we have proposed Multi-Objective Optimum of BPNCC (MOO-BPNCC) approach that consists of two independent and alternative methods (goal-base and minimization, maximization). We applied heuristics on these two independent methods to find the optimum solutions and found that the goal-base is infeasible due to consuming more space and execution time. The minimization and maximization is best as it removes optional features that are not required for optimization. In experiments, we calculated the outcomes of all three processes that show the significant improvement of optimum solution without constraint violation occurrence. We conduct experiment and results prove that our proposed optimization approach is better than previously proposed optimization algorithms for feature model, such as a non-dominated sorting genetic algorithm (NSGA-II) and an indicator-based evolutionary algorithm (IBEA). We apply MOO-BPNCC approach on same feature models that have optimized by using IBEA in previous research and found 100% correctness optimized feature configurations.