Utilizing Imprecise Results in Real-Time Systems

Abstract

In a hard real-time system, all real-time tasks must complete before their individual deadlines. Since the execution times of real-time tasks may vary and the load level on a system may fluctuate, it is difficult to guarantee that all deadlines are met at all times. In our model, a result produced by a computation is said to be precise if the computation runs to normal termination. If an intermediate result is made available by the computation at the deadline even when it has not totally completed, the result is said to be imprecise. This project is concerned with the design and evaluation of programming primitives, scheduling algorithms and system supports for the imprecise computation model. The model allows early results produced by partially completed computations to be effectively used to ensure the timeliness of real-time systems and to achieve graceful degradation. In the past year, we have looked into four issues related to real-time imprecise computations: timing constraint implementation, performance analysis, performance polymorphism, and scheduling parallelizable imprecise computations. In this report, we present the summary of research in each of these topics.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Oct 01, 1990
Accession Number
ADA253333

Entities

People

  • Jane W. Liu
  • Kwei-jay Lin

Organizations

  • University of Illinois Urbana–Champaign

Tags

DTIC Thesaurus Topics

  • Algorithms
  • Computations
  • Computer Languages
  • Computer Programming
  • Computer Programs
  • Computer Science
  • Computers
  • Electronic Mail
  • Language
  • Linear Programming
  • Models
  • Object Code
  • Parallel Computing
  • Parallel Processing
  • Programming Languages
  • Scheduling (Production)
  • Time

Fields of Study

  • Computer science

Readers

  • Educational Psychology
  • Neural Network Machine Learning.
  • Parallel and Distributed Computing.