Architecture-level configuration of industrial control systems: Foundations for an efficient approach

摘要

Configuration is a recurring problem in many domains. In an earlier work, we focused on architecture-level configuration of large-scale embedded software systems, in particular industrial control systems, and proposed a methodology that enables engineers to configure products by instantiating a given reference architecture model. Products have to satisfy a number of constraints specified in the reference architecture model. If not, the engineers have to backtrack their configuration decisions to rebuild a configured product that satisfies the constraints. Backtracking configuration decisions makes the configuration process considerably slow. In this paper, we improve our earlier work and propose a backtrack-free configuration mechanism. Specifically, we propose an algorithm that computes an ordering over configuration parameters that, for any cycle-free reference architecture model, yields a consistent configuration without any need to backtrack. We provide formal specification and proofs of termination, correctness, and completeness of our algorithm. We demonstrate the effectiveness of our approach using a simplified industrial case study. Results of our experiments show that our ordering approach eliminates backtracking in practice. It reduces the overall configuration time by reducing both the required number of value assignments, and the time that it takes to complete one configuration iteration. Moreover, we show that the latter has a linear growth with the size of the configuration problem.