Service Improvements in Real-Time Uniprocessor Scheduling With Single Errors

Abstract

Mixed-criticality scheduling in modern deeply embedded mission and safety-critical systems needs to consider delivered service, that is, the runtime in low criticality mode. If the change into a higher criticality mode is triggered by the first overrunning job, the service is severely reduced. With earliest deadline first with virtual deadlines for single errors (EDF-VD-SE) we show how to reserve additional time to tolerate a single overrunning job by formulating and solving an optimization problem, and that EDF-VD-SE is feasible without assumptions about error probabilities for safety guarantees. We conduct extensive simulation experiments to report on average doubled service figures, and show how EDF-VD-SE results in a nearly constant acceptance rate of random task systems.

Implications of Non-Uniform Deadline Scaling to Quality of Service Under Single Errors

“…In our anticipated applications WCET parameter overruns are expected [16]. As traditional dual-criticality scheduling discards low criticality tasks upon the first overrun, under the assumption that overruns are unlikely, we propose the following extensions for single overrun tolerance to provide better service to low criticality tasks.…”

“…During system operation, our mode-switched EDF scheduler is in one of three modes, as shown in Fig. 6 deadline first with virtual deadlines for single errors (EDF-VD-SE) [16]: During the initial low criticality mode, all tasks are allowed to generate jobs. The jobs arrive at the scheduler's queue, and are EDF-scheduled according to their absolute deadlines or virtual absolute deadlines: Jobs from low criticality tasks are scheduled according to their absolute deadlines, and jobs from high criticality tasks are scheduled according to their virtual absolute deadlines.…”

“…For our investigation we resort to random task sets, generated with the UUnifast algorithm [24], additional pessimism [16], and a mostly EDF-VD schedulable parameterization [25]: Periods are uniformly drawn between 𝑝 𝑙 = 50 and 𝑝 𝑢 = 200, and pessimism is uniformly drawn between 𝑧 𝑙 = 1 and 𝑧 𝑢 = 2. For each utilization in low criticality mode, we calculate the acceptance rate from 1024 task sets.…”

“…Nevertheless, in terms of acceptance rate all approaches are sensitive to the composition of their task sets. Therefore, detailed descriptions about random task set generation [16] are a necessity for comparable results.…”

“…Moreover, the lack of resources in deeply embedded systems rules out dynamic slack monitoring approaches [13] , or to precompute proper schedules for all scenarios at design time [14]. Therefore, approaches which emphasize static design time decisions over elaborate dynamic decisions during system operation are preferable, but verification-friendly state-of-the-art approaches, which resort to static slack analysis during design time [15], [16] , suffer from low acceptance rates. Hence, there is a strong need for a mixed-criticality approach for industrial fault-tolerant systems which provides degraded service to low criticality tasks, is sensible for verification, and likely finds a solution for highly loaded systems.…”