What is predictability for real-time systems?

Abstract

In summary, predictability in real-time systems has been defined in many ways. For static real-time systems we can predict the overall system performance over large time frames (even over the life of the system) as well as predict the performance of individual tasks. If the prediction is that 100% of all tasks over the entire life of the system will meet their deadlines, then the system is predictable without resorting to any stochastic evaluation. In dynamic real-time systems we must resort to a stochastic evaluation for part of the performance evaluation. Predictability for these systems should mean that we are able to satisfy the timing requirements of critical tasks with 100% guarantee over the life of the system, be able to assess overall system performance over various time frames (a stochastic evaluation), and be able to assess individual task and task group performance at different times and as a function of the current system state. If all these assessments meet the timing requirements, then the system is predictable with respect to its timing requirements.