On Fixed Priority Preemptive Scheduling for Imprecise Computation

We present new results in fixed priority scheduling for imprecise computation in real-time systems. Imprecise computation has not been popularized enough for reasons which are rooted from the task model, scheduling methods (EDF-based) and general system support. Audsley et al. have devised schedulability tests for guaranteeing mandatory computation while incorporating unbounded optional components. In particular, the task model is defined to be of the form “Prologue-Optional-Epilogue”, which is a generalisation of the traditional “MandatoryOptional” task model of Liu el al.. However, problems are discovered in the proposed schedulability tests which can make calculation of task’s worst-case response time inaccurate. Besides presenting the problem and proposing the required fixes, we prove the tests can be simplified under deadline monotonic priority ordering. We also propose a novel scheduling scheme, based on Dual Priority Scheduling, to maximize the scheduling of unbounded optional components, along with the required schedulability tests. The application of the new scheme shows substantial improvement. With its basis on fixed priority scheduling, it is expected the scheme can be easily incorporated into existing real-time operating systems and provoke wider use of imprecise computing.