Iterative design space exploration for networks requiring performance guarantees

Networks are embedded many system nowadays. They route control messages, sensory data, etc. and thus their functionality is integral to the enclosing system. In case of safety-critical systems, additional non-functional demands need to be fulfilled. A prime example are systems from the avionics domain where worst-case message delay needs to be bounded. The Deterministic Network Calculus (DNC) analysis is able to provide worst-case bounds on message delay. Accuracy of these bounds has been steadily improved to counteract costly over-provisioning. In this paper, we first provide detailed insight on the aspect preceding the DNC analysis: the network modeling step. It already contributes pessimism resulting in inaccuracy of later results. Second, we contribute multiple algorithms for iterative design space exploration within the limits of modeling that employ the most accurate DNC analysis. In numerical experiments, we show their effectiveness to reduce an avionics network's tail latency by optimized resources usage and the required effort.