Analysis of a simple model for non-preemptive blocking-free scheduling

Abstract

Non-preemptive scheduling is known for its lower efficiency in meeting temporal constraints when compared to preemptive scheduling. However, it is still used in certain cases such as in message scheduling over serial broadcast buses and in light multi-tasking kernels for embedded systems based on simple microprocessors. These cases are typically found in control applications requiring the periodic execution (or transmission) of a set of tasks (or messages) with low jitter. This paper refers to a simple execution strategy based on synchronised timetriggering and non-preemptive scheduling that allows to eliminate the blocking factor commonly associated to non-preemption and thus reduce activation jitter. The elimination of such blocking factor is achieved by using inserted idle-time. The paper focuses on the schedulability analysis of a generic task set executed according to the referred model. In one part, a specific response time-based analysis is presented which supports, under worst-case assumptions, a necessary and sufficient schedulability assessment. In a following part, the paper presents a general theorem that allows to adapt the existing analysis for preemptive scheduling to the referred model. In particular, this theorem allows to develop adequate utilization bounds for guaranteed schedulability, based on the well known bounds for ratemonotonic analysis.