Evaluating Mixed Criticality Scheduling Algorithms with Realistic Workloads

Most work on mixed-criticality scheduling has considered timing-related failures to be independent of one another. In reality this is not true as in many systems the state that caused the original failure will be similar to the state in the next release (job) of the task. Therefore when arguing about the number of jobs that do not meet their deadlines, it is crucial tasks have an appropriate fault model incorporated into the tool framework (i.e. task set generators and simulators) used to evaluate scheduling policies. The second issue that affects the tool framework is the choice of Worst-Case Execution Times (WCET) for different criticality modes of tasks. In the current literature it has been argued that a WCET should be chosen which would only be exceeded incredibly rarely, e.g. 1 in 1016 jobs. This leads to WCET values much greater than the High WaterMark (HWM). The needs of certification and the consideration of how safety is argued leads to the conclusion that the probability of a job not meeting its deadline can be much greater. This would greatly impact the WCETs and hence the results of the evaluation. The contributions of this paper are thus a more realistic tool framework, and hence more realistic results than those previously reported, which we claim gives a better insight into how the scheduling policies would behave in practice and hence better evidence for any safety case.