Guided Rule-Based Multi-objective Optimization for Real-Time Distributed Systems

Automated optimization of real-time architectures with respect to cost, performance, robustness and safety has received considerable attention in the last decade. In this paper, we present an automated Design Space Exploration (DSE)method based on both a multi-objective genetic algorithm and a heuristic particle-swarm-optimization technique. The optimization process is guided to desired solutions by weight coefficients that are assigned to the system objectives. The proposed method automatically generates architecture alternatives by changing hardware topology and mapping the tasks on different nodes, CPUs and by modifying their execution priority. Based on multiple quality objectives, the optimization method concludes to the Pareto-optimal solution set of the architecture alternatives. Moreover, in this paper we present an addition to the pre-existing optimization heuristics, targeting the reduction of the exploration time and maintaining a high-quality Pareto-optimal solution set. Finally, we compare the NSGA-II algorithm against the OMOPSO and their "Rule-Based Initial Population versions", for the convergence speed and the quality of their solutions, by comparing the hyper volume and the epsilon quality indicators. The proposed DSE approach has been applied to an autonomously navigating robot system consisting of several processing nodes (real-time distributed system) and resulting into better optimized and balanced solutions when compared to the proposed system architecture by an architect specialist