TERAFLUX: Exploiting Dataflow Parallelism in Teradevice Computing


    Motivation: Non-efficient exploitation of a large-scale of parallelism, by current programing models.
    TERAFLUX:Develop a coarse grain dataflow solution to harness large-scale parallelism.
    TERAFLUX tool chain targets programing models, compiler analysis, and a reliable architecture.

    University of Delaware Objectives in TERAFLUX

  • Survey state-of-the-art models in dataflow-inspired execution models and implementations.
  • Propose new methods/tools applicable to dataflow-inspired teradevices.
  • Evaluate the opportunity to add percolation and power-aware task scheduling directly using DF-thread.
  • Evaluate the opportunity to add percolation and power-aware scheduling to the Codelet Model.
  • Survey state-of-the-art techniques in code and data movement and power-aware task scheduling.
  • Comparison between the Codelet Execution Model and the TERAFLUX model: DF-thread.
  • Characterization of the differences between both the Codelet and DF-thread Models.
  • Joint research papers between the university of Delaware and the university of Siena.








© CAPSL 1996-2013. All Rights Reserved.