Improvement to Semi-Partitioned Cyclic Executives for Mixed-Criticality Scheduling on Multiprocessor Platforms

• Published in: IEEE access Vol. 8; pp. 223606 - 223617
• Main Author: Zhang, Fengxiang
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2020; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; cyclic executives; Mathematical model; Microprocessors; mixed-criticality systems; multi-core/single-chip multiprocessors; Multiprocessing; Optimization; Processor scheduling; Program processors; real-time and embedded systems; schedulability analysis; Schedules; Scheduling; Scheduling and task partitioning; semi-partitioned scheduling; Switches; Task analysis; Timing
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2020.3044724

This paper focuses on semi-partitioned cyclic executives for mixed-criticality multiprocessor systems in the case of which a job can be split and execute on different processors, this strategy incorporates most of the advantages of the fully partitioned scheduling while further maximising the total processor utilization. We propose algorithms in order to improve the schedulability of such a system. The length of each frame can be reduced to an optimal (minimum) value, which is a necessary schedulability condition for any schedule under this system model (i.e. every processor's capacity is fully utilized within each frame so that the frame cannot be further shortened), whilst all timing requirements of the system are satisfied. By conducting extensive experiments on a very large number of randomly generated job sets, it is demonstrated that our proposed algorithms significantly reduce the required length of each frame; it is possible for almost 100% of the job sets in the experiments to obtain their optimal values with the exception of a very limited number of specific situations. We also prove that when the complete times of some specific jobs are fixed given values, the strategy to shorten the frame presented in this paper is optimal.