Abstract: We consider a simulation tool for the study of scheduling algorithms for real-time systems composed of periodic and aperiodic tasks. We first introduce the scheduling problem of real-time systems, we pre\-sent popular scheduling algorithms for periodic tasks with hard deadlines, and then we consider the case of mix systems, composed of periodic as well aperiodic tasks and we briefly present the various popular scheduling algorithms for these kind of systems. After that, we describe our simulation tool and define the metrics we use to compare the performances of the various servers. In a next section, we present results of simulations and classify the aperiodic servers. In this section, we also present strategies to improve the performances of the various scheduling algorithms. Finally, we give some concluding remarks.
Abstract: This paper presents a method used in order to generate arbitrary task systems, compound of periodic and independent tasks. The originality of this methods relays on a judicious choice of the periods of the tasks in order to reduce the lcm of the periods (and the simulation duration required in order to study such a system). The authors prove that the lcm of n integers might be very high: they propose an upper bound of the lcm of n integers, given by exp(m) where m is the least prime number greater than the greatest integer in the set of the n integers. Then, they present a method in order to choose the periods of the tasks which imply a bounded lcm of the periods. The method relies on the decomposition of an integer in distinct prime factors. A second algorithm is proposed in order to obtain the other temporal parameters of the task system to generate. Then, an example is given in order to illustrate the method: several task systems are generated for performance analysis of classical scheduling algorithms, in term of preemption.
Personnal Informations
