RUN enhancement through Bayesian networks. Issue 5 (3rd September 2019)

• Record Type: Journal Article
• Authors: Hermosillo-Gomez, J. Angel; Benitez-Perez, Hector
• Is Part Of: International journal of parallel, emergent and distributed systems. Volume 34:Issue 5(2019)
• Page Start: 523
• Page End: 537
• Publication Date: 2019-09-03
• Subjects: Real-time -- Bayesian networks -- reduction to UNiprocessor; Parallel computers -- Periodicals; Electronic data processing -- Distributed processing -- Periodicals; Computer algorithms -- Periodicals; 004.35
• DOI: 10.1080/17445760.2019.1578880
• Languages: English
• ISSNs: 1744-5760

Abstract

ABSTRACT: This work is inspired by the Reduction to UNiprocessor (RUN) algorithm, wich schedules a set of periodic tasks considering their worst-case execution time (WCET). Even if the tasks are executed in less time than their WCETs, there is no possibility of rearranging the rest of the tasks to use the idle slots made available during execution. Therefore, the resulting problem consists of proposing a strategy to exploit the idle intervals. The difficulty of this problem increases when considering that the tasks that are accommodated first have stochastic execution times and that aperiodic tasks used to occupy the idle slots are composed of jobs that have precedence constraints. Besides, these jobs have stochastic temporal restrictions that must be met. To address this issue, a novel heuristic method based on Bayes networks is proposed. Two main contributions are presented. First is the mapping of the problem into a Bayesian network, and second is a heuristic that helps identify the best job, from the new set of aperiodic tasks, for occupying an idle slot created by RUN during its execution. There are two main goals in this work: to reduce global idle time and to execute any reliable job-shop by its deadline.