An on-line scheduling policy for IRIS real-time composite tasks

Abstract

In a new class of real-time tasks, called increasing reward with increasing service tasks (IRIS), the value of a task's computation increases as a function of the amount of execution time it has been able to accrue before its deadline expires. A composite task is a set of dependent component tasks. This paper presents an on-line scheduling policy for IRIS composite tasks. This policy aims to increase the total accrued reward of composite tasks within deadlines on a uniprocessor. If there is no enough time to schedule all the component tasks of a composite task, those component tasks having the little impact on the total reward of the composite task are not scheduled for execution by the proposed scheduling policy. A flexible task graph is also introduced to aid in determining the amount of service times to be allocated to the given composite tasks. The performance of the proposed scheduling policy is evaluated both analytically and by computer simulation.