Model-Based Representations for Dataflow Schedules

Shuvra S. Bhattacharyya*, Johan Lilius

*Korresponderande författare för detta arbete

Forskningsoutput: Kapitel i bok/konferenshandlingKapitelVetenskapligPeer review

2 Citeringar (Scopus)

Sammanfattning

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.

OriginalspråkEngelska
Titel på gästpublikationPrinciples of Modeling
Undertitel på gästpublikationEssays Dedicated to Edward A. Lee on the Occasion of His 60th Birthday
FörlagSpringer Verlag
Sidor88-105
Antal sidor18
ISBN (tryckt)978-3-319-95245-1
DOI
StatusPublicerad - 2018
MoE-publikationstypA3 Del av bok eller annan forskningsbok

Publikationsserier

NamnLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volym10760 LNCS
ISSN (tryckt)0302-9743
ISSN (elektroniskt)1611-3349
Namn Programming and Software Engineering
FörlagSpringer

Fingeravtryck

Fördjupa i forskningsämnen för ”Model-Based Representations for Dataflow Schedules”. Tillsammans bildar de ett unikt fingeravtryck.

Citera det här