• DocumentCode
    3590663
  • Title

    Execution trace graph analysis of dataflow programs: Bounded buffer scheduling and deadlock recovery using model predictive control

  • Author

    Casale-Brunet, S. ; Bezati, E. ; Mattavelli, M. ; Canale, M. ; Janneck, J.W.

  • Author_Institution
    EPFL SCI-STI-MM, Ecole Polytech. Fed. de Lausanne, Lausanne, Switzerland
  • fYear
    2014
  • Firstpage
    1
  • Lastpage
    6
  • Abstract
    Execution trace graph analysis of dataflow programs has been demonstrated to be an effective way for exploring and optimizing the design space of many core applications. In this work a novel transformation from the execution trace graph to an event driven linear system is proposed. It is also illustrated how the trace space of can be effectively reduced and well known system control techniques can be efficiently used in order to find close to optimal solutions. In particular, the problem of finding a bounded buffer size configuration is proposed and solved using a model predictive controller. Two design examples, a JPEG and an MPEG HEVC decoder have been used to demonstrate the effectiveness of the approach.
  • Keywords
    data flow computing; graph theory; optimisation; predictive control; scheduling; JPEG; MPEG HEVC decoder; bounded buffer scheduling; dataflow programs; deadlock recovery; event driven linear system; execution trace graph analysis; model predictive control; Algorithm design and analysis; Heuristic algorithms; Optimization; Petri nets; Predictive control; System recovery; Transform coding;
  • fLanguage
    English
  • Publisher
    ieee
  • Conference_Titel
    Design and Architectures for Signal and Image Processing (DASIP), 2014 Conference on
  • Type

    conf

  • DOI
    10.1109/DASIP.2014.7115623
  • Filename
    7115623