Research On Scheduling Method For Mixed-Criticality Tasks In Multicore Systems

With the emergence of multi-core processors,the processing performance of computers has greatly improved,leading to higher requirements for embedded systems.In order to save space,costs,power consumption,and computational resources of embedded systems,multiple functions or subsystems with different criticality levels are integrated into the same hardware platform,forming mixed-criticality systems.In such systems,multiple real-time periodic tasks of various critical levels are executed simultaneously.Due to the potential catastrophic consequences of failure in high criticality tasks,while failure in low criticality tasks only results in a decrease in user experience,most of the current research on mixed-criticality task scheduling algorithms tends to delay or suspend a portion of low criticality tasks to ensure the execution requirements of high criticality tasks.This conservative approach leads to lower resource utilization,making the effective partitioning of tasks to fully utilize the performance of multi-core processors a focal point of research in mixed-criticality systems.To address the aforementioned issues,this thesis aims to ensure the execution of high criticality tasks while improving the passive handling of low criticality tasks in mixed-criticality systems,thereby enhancing the utilization of individual processors and improving overall system efficiency.The main contributions of this thesis are as follows:（1）A mixed-criticality task global scheduling method based on comprehensive impact factors is proposed to address the passive handling of low criticality tasks in mixed-criticality systems.By considering the relationships between various task attributes and defining several factors that influence task priority determination,a novel approach for determining task priorities is introduced.The correctness of this method is theoretically derived and experimentally validated.Results demonstrate that compared to existing classical scheduling algorithms,this algorithm improves the schedulability of low criticality tasks by 10%and reduces task migration and preemption overheads by 17.9%.（2）To address the issue of high task migration overhead in the mixed-criticality task global scheduling algorithm,an adaptation of the traditional semi-partitioned scheduling algorithm EDF-os（Earliest Deadline First Based Optimal Semi-partitioned）is proposed for mixed-criticality systems,named MC-EDF-os（Mixed-Criticality Earliest Deadline First Based Optimal Semi-partitioned）.During task execution,when the system criticality level changes,a small portion of low criticality tasks is allowed to migrate to lightly loaded processors for continued execution.Experimental results show that compared to the EDF-VD（Earliest Deadline First Virtual Deadline）algorithm,this algorithm reduces migration overhead by 38%and improves the schedulability of low criticality tasks by 14.8%,while maintaining a more balanced utilization of individual processors.（3）Most of the current mixed-criticality task scheduling algorithms are verified through simulation experiments and lack validation of effectiveness and correctness in real operating systems.This thesis implements the mixed-criticality task scheduling algorithm based on comprehensive impact factors and its comparative algorithms on the LITMUS^RT real-time validation platform,integrating them as plugins into the platform kernel,and finally validating them through instances.The results demonstrate that the proposed algorithm outperforms the CAPA（Criticality as Priority Assignment）algorithm and FBTWP-CFP（Forward and Backward Time Window Partition Criticality Factor Prior）algorithm in the practical validation platform,thereby proving its effectiveness and correctness.