Probabilistic Real-Time Data Access with Interval Constraints

Real-time data management for manufacturing control applications indicates a need for probabilistic deadline guarantees and interval constraints. When shared data objects are treated as resources, real-time data management becomes a resourceconstrained real-time scheduling problem. In this paper, we perform simulations and measurements to evaluate the utilities of existing real-time scheduling algorithms in our University of Michigan Open-Architecture Controller testbed environment. We exhibit that, among rate-monotonic, first-infirst-out and earliest-deadline-first, none can consistently outperform the others in terms of miss ratio. To satisfy interval constraints in the presence of timer interval variance and “memory,” resetting timers is intuitively helpful. However, we find that it is not a viable approach even when the associated overhead is negligible. Rate-monotonic and first-in-first-out perform well in simulations, but the real-time operating system (QNX) has significant unpredictability—thus the performance is not as satisfactory in the testbed. To our knowledge, this research is the first empirical studies of real-time scheduling algorithms in the context of probabilistic deadline guarantees and interval constraints.