Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

Dataflow Programs Analysis and Optimization Using Model Predictive Control Techniques : Two Examples of Bounded Buffer Scheduling: Deadlock Avoidance and Deadlock Recovery Strategies

Canale, M. ; Casale-Brunet, S. ; Bezati, E. ; Mattavelli, M. and Janneck, J. LU (2016) In Journal of Signal Processing Systems 84(3). p.371-381
Abstract

The analysis of the trace graphs generated by dataflow program executions has been shown to be an effective tool for exploring and optimizing the design space of application programs on manycore/multicore platforms. In this work a new approach aiming at finding bounded buffer size configurations for implementations generated by dataflow programs is presented. The introduced method is based on an original transformation procedure which converts the execution trace graph into an event driven linear system made up by a Petri Net. A control theoretic approach based on Model Predictive Control methodologies is then applied to the obtained Petri Net system in order to effectively explore the dataflow program design space and find nearly... (More)

The analysis of the trace graphs generated by dataflow program executions has been shown to be an effective tool for exploring and optimizing the design space of application programs on manycore/multicore platforms. In this work a new approach aiming at finding bounded buffer size configurations for implementations generated by dataflow programs is presented. The introduced method is based on an original transformation procedure which converts the execution trace graph into an event driven linear system made up by a Petri Net. A control theoretic approach based on Model Predictive Control methodologies is then applied to the obtained Petri Net system in order to effectively explore the dataflow program design space and find nearly optimal buffer dimensioning solutions leading to a deadlock free program execution. Two real challenging design case examples, namely a JPEG and a MPEG HEVC decoder, are introduced to show the effectiveness of the introduced approach.

(Less)
Please use this url to cite or link to this publication:
author
; ; ; and
organization
publishing date
type
Contribution to journal
publication status
published
subject
keywords
Buffer minimization, Causation trace graph, Design space exploration, Event driven system, Model predictive control, Petri net
in
Journal of Signal Processing Systems
volume
84
issue
3
pages
11 pages
publisher
Springer
external identifiers
  • wos:000379010700007
  • scopus:84946780948
ISSN
1939-8018
DOI
10.1007/s11265-015-1083-4
language
English
LU publication?
yes
id
499a8a38-f248-4961-b9d4-6b73744c47f5
date added to LUP
2016-12-13 08:26:30
date last changed
2024-01-04 18:53:09
@article{499a8a38-f248-4961-b9d4-6b73744c47f5,
  abstract     = {{<p>The analysis of the trace graphs generated by dataflow program executions has been shown to be an effective tool for exploring and optimizing the design space of application programs on manycore/multicore platforms. In this work a new approach aiming at finding bounded buffer size configurations for implementations generated by dataflow programs is presented. The introduced method is based on an original transformation procedure which converts the execution trace graph into an event driven linear system made up by a Petri Net. A control theoretic approach based on Model Predictive Control methodologies is then applied to the obtained Petri Net system in order to effectively explore the dataflow program design space and find nearly optimal buffer dimensioning solutions leading to a deadlock free program execution. Two real challenging design case examples, namely a JPEG and a MPEG HEVC decoder, are introduced to show the effectiveness of the introduced approach.</p>}},
  author       = {{Canale, M. and Casale-Brunet, S. and Bezati, E. and Mattavelli, M. and Janneck, J.}},
  issn         = {{1939-8018}},
  keywords     = {{Buffer minimization; Causation trace graph; Design space exploration; Event driven system; Model predictive control; Petri net}},
  language     = {{eng}},
  month        = {{09}},
  number       = {{3}},
  pages        = {{371--381}},
  publisher    = {{Springer}},
  series       = {{Journal of Signal Processing Systems}},
  title        = {{Dataflow Programs Analysis and Optimization Using Model Predictive Control Techniques : Two Examples of Bounded Buffer Scheduling: Deadlock Avoidance and Deadlock Recovery Strategies}},
  url          = {{http://dx.doi.org/10.1007/s11265-015-1083-4}},
  doi          = {{10.1007/s11265-015-1083-4}},
  volume       = {{84}},
  year         = {{2016}},
}