Communication Dans Un Congrès Année : 2025

Communication and Shared Memory Efficient Mapping Techniques of Real-Time DAGs upon Clustered Multicore Platforms

Résumé

When it comes to mapping and scheduling Real-Time applications on parallel platforms, the Directed Acyclic Graphs (DAGs) processing model is commonly used. In this paper, we leverage the structural properties of DAG-based applications to optimize their deployment in clustered multicore processors. We assume a constant communication time between clusters and partitioned banked memory inside clusters. Our main contribution is a novel mapping (and ordering) heuristic that accounts for both communication size and DAG topology, while respecting memory constraints. Compared to the classic list scheduling-based mapping algorithm HLFET, when our approach succeeds in finding a mapping, it achieves a reduction of up to 75% in the global DAG response time and leads to better scheduling in 80% of the cases.

Fichier principal
Vignette du fichier
map-sched-preprint.pdf (951.81 Ko) Télécharger le fichier

Dates et versions

hal-05289989 , version 1 (18-02-2026)

Licence

Identifiants

  • HAL Id : hal-05289989 , version 1

Citer

Matheus Schuh, Claire Maiza, Pascal Raymond, Bruno Ferres, Joël Goossens, et al.. Communication and Shared Memory Efficient Mapping Techniques of Real-Time DAGs upon Clustered Multicore Platforms. 33rd International Conference on Real-Time Networks and Systems, Nov 2025, Pise, Italy. ⟨hal-05289989⟩
231 Consultations
32 Téléchargements

Partager

  • More