Graph-to-Segment Transformation Technique minimizing the number of processors for Real-time Multiprocessor Systems

Manar Qamhieh (1) , Serge Midonnet (1) , Laurent George (1)

1 LIGM - Laboratoire d'Informatique Gaspard-Monge

In energy-aware real-time systems, there are techniques to optimize the energy consumption through turning off idle processors, Dynamic Power Management (DPM) is an example of such techniques. In this paper we propose a transformation technique of graph task model into the multi-thread segment task minimizing the number of processors needed to schedule the tasks. This reduces the energy consumption of the system when DPM like technique is used. The Directed Acyclic Graph task is presented as a directed graph of subtasks under precedence constraints, while the other task is called the multi-threaded segment model, in which a task is a sequence of segments, and each segment has a number of threads and an intermediate deadline. The graph model is more general than the segment model, but it is more complicated to schedule and analyze, because in the segment model, the threads of each segment are scheduled as independent sequential tasks on multiprocessor platform. Due to the dependencies between the subtasks of a graph, it could have a number of possible segment combinations. In this paper, we propose a graph-to- segment transformation technique, which generates a multi- thread segment task with minimum number of processors required to execute on multiprocessor platform.

Mots clés: Graphs; Real-time scheduling; Parallel systems; Multiprocesessor systems; Dynamic power managment

Domaines: Calcul parallèle, distribué et partagé [cs.DC]; Systèmes embarqués; Algorithme et structure de données [cs.DS]

Manar Qamhieh, Serge Midonnet, Laurent George. Graph-to-Segment Transformation Technique minimizing the number of processors for Real-time Multiprocessor Systems. Workshop on Power, Energy, and Temperature Aware Real-time Systems (PETARS), Dec 2012, San Juan, Puerto Rico. ⟨hal-00846942⟩