Exploiting Approximations in Real-Time Scheduling

Kamyar Mirzazad Barijough, Lin Huang, I. Hong Hou, Sachin S. Sapatnekar, Jiang Hu, Andreas Gerstlauer

Research output: Chapter in Book/Report/Conference proceedingChapter

Abstract

In real-time systems, to provide timing guarantees, pessimistic worst-case bounds are traditionally assumed for computation and communication tasks. In practice, tighter bounds can be established by allowing computation or communication to be dropped or to execute in an approximate or imprecise manner. This creates a fundamental tradeoff between tightness of bounds and degradations in application quality. In this chapter, we present scheduling of computation and communication tasks as a quality optimization problem in terms of computation and communication budget assignments for systems with independent and dependent tasks. For independent tasks, traditional mixed-criticality (MC) systems can be extended to derive the precision of low-criticality tasks such that combined quality degradation is minimized while satisfying schedule admissibility. For dependent tasks, we describe approaches to find an optimized mapping, scheduling, and budgeting of task graphs that maximizes overall quality while meeting end-to-end real-time constraints. We evaluate our proposed approaches on both artificial and real-world task sets and compare them to traditional solutions that do not allow for approximations.

Original languageEnglish (US)
Title of host publicationApproximate Computing Techniques
Subtitle of host publicationFrom Component- to Application-Level
PublisherSpringer International Publishing
Pages287-322
Number of pages36
ISBN (Electronic)9783030947057
ISBN (Print)9783030947040
DOIs
StatePublished - Jan 1 2022

Bibliographical note

Publisher Copyright:
© Springer Nature Switzerland AG 2022.

Keywords

  • Approximate computing
  • High performance
  • Imprecise computing
  • Low energy
  • Low power
  • Mixed-criticality systems
  • Multiprocessor systems
  • Real-time systems
  • Task mapping
  • Task scheduling

Fingerprint

Dive into the research topics of 'Exploiting Approximations in Real-Time Scheduling'. Together they form a unique fingerprint.

Cite this