Automating Hardware and Software Evolution Analysis

Abstract

Cost-effective software evolution is critical to many distributed real-time and embedded (DRE) systems. Selecting the lowest cost set of software components that meet DRE system resource constraints, such as total memory and available CPU cycles, is an NP-Hard problem. This paper provides three contributions to R\&D on evolving software-intensive DRE systems. First, we present the Software Evolution Analysis with Resources (SEAR) technique that transforms component-based DRE system evolution alternatives into multidimensional multiple-choice knapsack problems. Second, we compare several techniques for solving these knapsack problems to determine valid, low-cost design configurations for resource constrained component-based DRE systems.\if false with multiple potential upgrades that cannot be solved efficiently by existing techniques\fi. Third, we empirically evaluate the techniques to determine their applicability in the context of common evolution scenarios. Based on these findings, we present a taxonomy of the solving techniques and the evolution scenarios that best suit each technique.