Rany Kahil, Peter Poplavko, Dario Socci, Saddek Bensalem
Predictability in Mixed-Criticality Systems (2018)
Predictability in Mixed-Criticality Systems (2018)
TR-2018-8.pdf
Keywords: mixed-critical systems, real-time scheduling, computational complexity
Abstract: Showing that a policy is predictable is a means to ensure sustainability of a solution proposed by a scheduling algorithm. A typical way to test a solution for correctness consists in conservative evaluation of its behavior under a specially selected set of high- workload runtime scenarios. Sustainability is a property that states that if a solution is deemed correct then it satisfies all the deadlines not only under the set of high- workload scenarios, but also in all less workload scenarios. The predictability property is a special case of sustainability under the assumptions that the workload is measured only by job execution times and that the correctness test consists of a simulation of the scheduling policy for finite set of jobs. In this paper we extend the predictability prop- erty to the case of mixed criticality jobs. We restrict ourselves for simplicity with just two levels of criticality. The ordinary, single criticality, scheduling policies commonly used in practice, are predictable in the usual sense, and thus it is sufficient to test them just for a single scenario with the worst-case execution times. We show that the typical generalization of the common scheduling policies to support the mixed criticality spec- ifications renders them non-predictable, whereas they are still predictable according to our proposed extended definition. We also show that our definition implies applica- bility of a previously existing method for testing mixed-criticality jobs by simulation, which previously had only a narrow application.
