Abstract |
---|
In parallel computing, performance is related both to algorithmic design choices, at the application level and to the scheduling strategy. Concerning dynamic scheduling, general classifications have been proposed that outline two fundamental units, related to control and information. However, those classifications are not precise enough to derive directly implementations that allow code reusability, providing facilities for the development of specific strategies. In this paper, we propose a generic modular specification, based on four components which interactions are precisely described. This specification has been used for implementation of various scheduling algorithms in two different parallel programming environments: PM2 (Espace) and Athapascan (Apache). |
Contact |
Gerson G. H. Cavalheiro LMC - IMAG,BP 53,100 rue des Mathematiques,38041 Grenoble CEDEX 9, FRANCE Gerson.Cavalheiro@imag.fr |