T. Ayav, G. Ferrari-Trecate, and S. Ylmaz. Stability properties of adaptive real-time feedback scheduling: A statistical approach. Proc. 12th International conference on real-time systems - RTS embedded systems , 2004. Paris, France, 30 March - 1 April.
This paper focuses on thestatistical analysis of an adaptive real-time feedback schedulingtechnique based on imprecise computation. We consider two-versiontasks made of a mandatory and an optional part to be scheduledaccording to a feedback control rate-monotonic algorithm. AProportional-Integral-Derivative (PID) control action provides thefeedback strategy for deciding about the execution or rejection ofthe optional sub-tasks. By modelling the task execution times asrandom variables, we compute the probability density of the CPUutilization and derive conditions on PID parameters guaranteeingthe stability of the overall system around a desired level of CPUutilization. This allows us to highlight the tasks statistics andthe scheduling parameters that affect critically stability. Theanalysis is developed by first exploiting a number of simplifyingassumptions that are progressively removed. The main results arealso demonstrated through monte-carlo simulations of thescheduling algorithm.