Tampere University of Technology

TUTCRIS Research Portal

Model-Based Representations for Dataflow Schedules

Research output: Chapter in Book/Report/Conference proceedingChapterScientificpeer-review

Details

Original languageEnglish
Title of host publicationPrinciples of Modeling
PublisherSpringer Verlag
Pages88-105
Number of pages18
ISBN (Electronic)978-3-319-95246-8
ISBN (Print)978-3-319-95245-1
DOIs
Publication statusPublished - 2018
Publication typeA3 Part of a book or another research book

Publication series

NameLecture Notes in Computer Science
Volume10760
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Abstract

Dataflow is widely used as a model of computation in many application domains, especially domains within the broad area of signal and information processing. The most common uses of dataflow techniques in these domains are in the modeling of application behavior and the design of specialized architectures. In this chapter, we discuss a different use of dataflow that involves its application as a formal model for scheduling applications onto architectures. Scheduling is a critical aspect of dataflow-based system design that impacts key metrics, including latency, throughput, buffer memory requirements, and energy efficiency. Deriving efficient and reliable schedules is an important and challenging problem that must be addressed in dataflow-based design flows. The concepts and methods reviewed in this chapter help to address this problem through model-based representations of schedules. These representations build on the separation of concerns between functional specification and scheduling in dataflow, and provide a useful new class of abstractions for designing dataflow graph schedules, as well as for managing, analyzing, and manipulating schedules within design tools.

Keywords

  • Dataflow, Model-based design, Scheduling, Signal processing

Publication forum classification

Field of science, Statistics Finland