Progressive and explicit refinement of scheduling for multidimensional data-flow applications using UML MARTE

Calin Glitia 1 Julien Deantoni 1 Frédéric Mallet 1 Jean-Vivien Millo 1 Pierre Boulet 2 Abdoulaye Gamatié 3
1 I3S - Laboratoire d'Informatique, Signaux, et Systèmes de Sophia Antipolis
2 LIFL - Laboratoire d'Informatique Fondamentale de Lille
3 ADAC - ADAptive Computing
LIRMM - Laboratoire d'Informatique de Robotique et de Microélectronique de Montpellier
Abstract : Process networks and data-flow graphs are used to capture data-dependencies in computation-intensive embedded systems. Their simplicity allows the computation of static schedules that reduce the dynamic overhead and increase predictability. The resulting schedule is a total ordering of actor computations and communications. It can therefore become an over-specification of the initial system when several schedules are valid. This is particularly the case for multidimensional data-flow applications. We propose a methodology to avoid such an over-specification. We propose to use logical time to capture explicitly all the valid schedules for a given multi-dimensional data-flow model. Then, we show that the proposed approach allows for a progressive and explicit refinement of computation scheduling that also captures constraints imposed by the environment and the execution platform. All this is achieved by using uml marte concepts and the resulting models can be considered for simulation and analysis with existing tools for early design validation. The whole approach is validated on a typical application devoted to radar signal processing.
Keywords : MARTE Clock constraint specification language Embedded system design Logical time Repetitive structure modeling Multidimensional data-flow
Type de document :
Article dans une revue
Design Automation for Embedded Systems, Springer Verlag, 2015, 19 (1-2), pp.1-33. 〈10.1007/s10617-014-9140-y〉
Liste complète des métadonnées
https://hal-lirmm.ccsd.cnrs.fr/lirmm-01912854
Contributeur : Abdoulaye Gamatié <>
Soumis le : lundi 5 novembre 2018 - 17:11:55
Dernière modification le : jeudi 8 novembre 2018 - 01:17:18

Identifiants

Collections

LIRMM | UNIV-MONTPELLIER | ADAC | I3S | UNIV-LILLE3 | LIRMM_MIC | MIC | MIPS | INRIA | UNICE | UNIV-COTEDAZUR

Citation

Calin Glitia, Julien Deantoni, Frédéric Mallet, Jean-Vivien Millo, Pierre Boulet, et al.. Progressive and explicit refinement of scheduling for multidimensional data-flow applications using UML MARTE. Design Automation for Embedded Systems, Springer Verlag, 2015, 19 (1-2), pp.1-33. 〈10.1007/s10617-014-9140-y〉. 〈lirmm-01912854〉

Exporter

BibTeX TEI DC DCterms EndNote

Partager

Métriques

Contact

support.ccsd.cnrs.fr
hal.support@ccsd.cnrs.fr
Logo CCSD