基于异构多核系统的实时混合关键性程序调度研究

Heterogeneous multicore systems are gradually becoming the mainstream computing platforms, but the mixed-criticality scheduling problem on heterogeneous multicore systems has not yet received enough attention. The approaches devised for mixed-criticality scheduling on homogeneous multicore systems are not directedly applicable to heterogeneous multicore systems. To address this issue, this project aims to investigate the scheduling problem of mixed-criticality applications on heterogeneous multicore systems. The main research targets are as follows: considering the different task model and the feature of heterogeneous multicore systems, we investigate the partitioned scheduling of mixed-criticality applications on heterogeneous multicore systems by means of different schedulability tests and propose new partitioned scheduling algorithms. In addition, in order to mitigate the resource wastage of the partitioned scheduling algorithms, we further investigate the semi-partitioned scheduling of mixed-criticality applications. By carefully considering the results obtained from the partitioned scheduling, the semi-partitioned scheduling of traditional real-time applications and the features of mixed-criticality applications, we propose a new semi-partitioned scheduling algorithm for scheduling mixed-criticality tasks on heterogeneous multicore systems. One novelty of this project is that besides the theoretical comparison between different scheduling algorithms, all proposed scheduling algorithms will be evaluated on a real-life platform in order to justify their applicability in practical cases. The outcome of this project will set up the theoretical foundation for executing mixed-criticality applications on heterogeneous multicore systems, and make a significant progress in the development of mixed-criticality systems on heterogeneous multicores.

异构多核系统已经逐渐成为主流的计算平台，然而基于异构多核系统的混合关键性程序调度还鲜有报道。现有的同构多核调度算法，不能够直接应用到异构多核系统上。针对这一问题，本项目拟研究基于异构多核系统的混合关键性程序调度问题，主要研究内容包括：针对不同的程序事件模型，研究基于不同可调度分析方法的异构多核系统划分调度算法，提出新的划分调度算法；针对划分调度算法可能存在的资源浪费情况，进一步研究半划分调度算法，基于划分调度的理论结果以及传统实时系统的半划分调度方法，考虑混合关键性程序的特点，设计适用于异构多核系统的混合关键性程序半划分调度。本课题另一个创新点在于，除了对调度算法的理论性能进行比较，课题组拟在一个真实平台实现所提调度算法，以验证其可行性。研究成果将为混合关键性程序在新型异构多核平台的执行提供了理论基础，积极推进异构多核混合关键性系统的发展。

为了追求更高的性能和更好的功耗比，嵌入式系统和边缘硬件的复杂度和异构性逐渐升高。于此同时程序也在逐渐变地多样化，尤其是是针对有实时需求和延时敏感的程序，而过去的10年中基于深度学习的智能程序成为了嵌入式系统和边缘设备上一个新的趋势。本项目立足于研究异构设备上的混合关键行调度问题，在三年的项目执行期内，从能耗，延迟，温度等多个方面深入研究了问题。在研究过程中，项目组积极追踪最新的研究动态，扩展项目的基础研究问题，提出了基于传统调度方法的能耗感知调度算法，基于强化学习的智能温度感知调度算法，同时也将问题进一步扩展将所提出的方法应用到了新的边缘人工智能系统中，进一步提出了基于边缘设备的持续学习和动态网络的研究。所提出的多种调度算法和优化方法，相比于现有方法都有着极大的提升。项目总计发表7篇高水平论文，其中IEEE Transactions 2篇，CCF B类会议1篇，C类会议1篇，其他均为EI或SCI检索，所提出的方法和发表论文收到了多方的关注。申请授权发明专利一项，并已经实现成果转换。本次项目顺利完成，为申请人后续的科研工作提供了极大帮助。

内容获取失败，请点击重试