Efficient Contention-Aware Scheduling of SDF Graphs on Shared Multi-bank Memory

Hai Nam Tran 1 Alexandre Honorat 2 Jean-Pierre Talpin 3 Thierry Gautier 3 Loïc Besnard 3
IBNM - Institut Brestois du Numérique et des Mathématiques, Lab-STICC - Laboratoire des sciences et techniques de l'information, de la communication et de la connaissance
3 TEA - Tim, Events and Architectures
Inria Rennes – Bretagne Atlantique , IRISA-D4 - LANGAGE ET GÉNIE LOGICIEL
Abstract : Novel memory architectures have been introduced in multi/many-core processors to address the performance bottle neck due to shared memory accesses. Taking the advantages brought by these architectures in scheduling analysis is still an open challenge. In this article, we present a scheduling analysis technique that exploits a shared multi-bank memory architecture to efficiently schedule parallel real-time applications modeled as synchronous data flow (SDF) graphs by minimizing the memory access contentions. Our approach aims at producing a static time-triggered schedule with the objective of minimizing the makespan and buffer size requirements while respecting consistency and data dependency constraints. An Integer Linear Programming formulation of the scheduling problem is presented, as well as a heuristic with significantly lower time complexity. Experimental results are given using synthetic SDF graphs generated by the SDF3 tool and applications available in the StreamIt benchmark.
Document type :
Conference papers
Complete list of metadatas

Cited literature [32 references]  Display  Hide  Download

Contributor : Jean-Pierre Talpin <>
Submitted on : Wednesday, July 24, 2019 - 3:48:44 PM
Last modification on : Friday, December 13, 2019 - 10:42:17 AM


Files produced by the author(s)


  • HAL Id : hal-02193639, version 1


Hai Nam Tran, Alexandre Honorat, Jean-Pierre Talpin, Thierry Gautier, Loïc Besnard. Efficient Contention-Aware Scheduling of SDF Graphs on Shared Multi-bank Memory. ICECCS 2019 - 24th International Conference on Engineering of Complex Computer Systems, Nov 2019, Hong Kong, China. pp.1-10. ⟨hal-02193639v1⟩



Record views


Files downloads