Scholarlink

<p>Mixed-criticality systems (MCS) integrate tasks with varying levels of criticality, making efficient scheduling essential for fault tolerance and real-time performance. While partitioned scheduling offers low overhead, traditional allocation heuristics often neglect runtime costs such as context switching, release latency, and scheduling overhead. The need for active backups of the tasks to achieve fault tolerance in safety-critical systems increases the overall system utilization and runtime overhead. The study examines the vital issue of timing analysis in safety-critical systems, highlighting the frequently neglected overheads in real-time operating systems. Also, there is a need for a scheduling framework in MCS with numerous goals, such as fault tolerance and overhead minimization. This paper proposes a task clustering-based (Clust) heuristic for partitioned earliest deadline first (P-EDF) scheduling, which groups tasks into heavy and light clusters to reduce overhead in fault-tolerant environments. Experiments conducted on Intel i7 HQ and ARM Cortex (R-Pi) platforms demonstrate that the proposed heuristic reduces context switch overhead by approximately 11.5% and scheduling overhead by up to 24% compared to utilization- and criticality-based allocations. The scheduling overhead is reduced in the proposed clust algorithm by around 10.0% on the Intel processor and by around 24.0% on the ARM Cortex processor. The work is done with the assumption that there is complete isolation of the processors and that there is no intercommunication interference.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Clustering-based task allocation for overhead reduction in multi-core mixed-critical systems

  • 期刊论文

  • 发表日期：

    2025年11月10日

查看全文
  • Preeti Godabole,
  • Aniket Samudre,
  • Sandeep S. Udmale,
  • Girish Bhole

摘要

Mixed-criticality systems (MCS) integrate tasks with varying levels of criticality, making efficient scheduling essential for fault tolerance and real-time performance. While partitioned scheduling offers low overhead, traditional allocation heuristics often neglect runtime costs such as context switching, release latency, and scheduling overhead. The need for active backups of the tasks to achieve fault tolerance in safety-critical systems increases the overall system utilization and runtime overhead. The study examines the vital issue of timing analysis in safety-critical systems, highlighting the frequently neglected overheads in real-time operating systems. Also, there is a need for a scheduling framework in MCS with numerous goals, such as fault tolerance and overhead minimization. This paper proposes a task clustering-based (Clust) heuristic for partitioned earliest deadline first (P-EDF) scheduling, which groups tasks into heavy and light clusters to reduce overhead in fault-tolerant environments. Experiments conducted on Intel i7 HQ and ARM Cortex (R-Pi) platforms demonstrate that the proposed heuristic reduces context switch overhead by approximately 11.5% and scheduling overhead by up to 24% compared to utilization- and criticality-based allocations. The scheduling overhead is reduced in the proposed clust algorithm by around 10.0% on the Intel processor and by around 24.0% on the ARM Cortex processor. The work is done with the assumption that there is complete isolation of the processors and that there is no intercommunication interference.