基于异构多核的静态任务调度策略研究

发布时间：2018-05-30 09:03

本文选题：异构多核处理器 + 静态任务调度　；参考：《哈尔滨工程大学》2012年硕士论文

【摘要】：从单核处理器时代开始，任务调度的研究就倍受众多专家、学者的关注，任务调度的结果直接影响操作系统的性能，系统性能的提升不仅要依靠自身的硬件水平，，还取决于加载在硬件之上的软件。随着多核处理器的出现，专家、学者将目光投向多核处理器上的任务调度，目前多核处理器上的任务调度已经成为高性能处理器研究热点之一。为了迎合处理器未来的发展趋势，本文旨在研究异构多核处理器上的任务调度，其目的在于寻找一种较理想的静态任务调度策略，使运行在异构多核处理器上的任务能得到合理的调度，缩短任务调度的总长度、提升多核处理器的执行效率。本文在对四种经典异构多核处理器任务调度算法分析和研究的基础上，针对现存高效任务调度算法存在的不足，提出一种基于加权优先级的静态任务调度策略，并设计了基于该策略的任务调度算法。本策略通过任务合并、任务分层和任务权值计算三个过程顺序执行来实现任务优先级计算阶段，然后通过任务任务分配到处理器和任务调度结果优化两个过程交替执行来实现任务映射到处理器阶段。新策略保证任务按层次顺序执行，并优先调度每一层的关键任务，尽量缩短关键路径的长度以减少总的任务完成时间。在调度过程中采用任务复制技术，以任务的冗余执行来提前后继任务的最早开始时间，并及时对冗余任务进行删除处理，缩短任务调度时间，提高多核处理器的并行性。为了验证基于加权优先级的任务调度策略的高效性和可行性，本文对提出的算法采用实验的方法进行验证，实验平台选用高性能的Simics模拟器，使用Simics模拟多核处理器的基本结构，并将新策略转化成具体的算法加载在模拟器上运行。实验结果表明：与以往的算法相比，新的算法继承了它们保证关键任务优先运行的优点，改善了优先级选择不当、冗余任务处理过晚的缺点，在任务调度过程中提高了处理器的并行性、减少了总的任务调度长度，达到算法改进的预期效果，具有一定的实际应用价值。
[Abstract]:Since the era of single core processor, the research of task scheduling has attracted the attention of many experts and scholars. The results of task scheduling directly affect the performance of the operating system, and the improvement of system performance depends not only on its own hardware level, but also on its own hardware level. It also depends on the software that is loaded on the hardware. With the emergence of multi-core processors, experts and scholars focus on task scheduling on multi-core processors. Task scheduling on multi-core processors has become one of the research hot spots in high-performance processors. In order to meet the future development trend of processors, this paper aims to study task scheduling on heterogeneous multi-core processors, which aims to find an ideal static task scheduling strategy. The tasks running on heterogeneous multi-core processors can be reasonably scheduled, the total length of task scheduling can be shortened, and the execution efficiency of multi-core processors can be improved. Based on the analysis and research of four classic heterogeneous multi-core processor task scheduling algorithms, this paper proposes a static task scheduling strategy based on weighted priority, aiming at the shortcomings of the existing efficient task scheduling algorithms. A task scheduling algorithm based on this strategy is designed. This strategy implements the task priority calculation phase by three sequential execution processes: task merging, task stratification and task weight calculation. Then the task mapping to processor phase is realized by alternate execution of task assignment to processor and task scheduling result optimization. The new strategy ensures that tasks are executed in hierarchical order, and prioritizes the scheduling of critical tasks in each layer, and minimizes the length of critical paths to reduce the total task completion time. In the process of scheduling, the task replication technology is adopted to advance the earliest start time of the subsequent task by redundant execution of the task, and to delete the redundant task in time to shorten the task scheduling time and improve the parallelism of the multi-core processor. In order to verify the efficiency and feasibility of the task scheduling strategy based on weighted priority, this paper uses the experimental method to verify the proposed algorithm. The experiment platform uses a high-performance Simics simulator. Simics is used to simulate the basic structure of multi-core processor, and the new strategy is transformed into concrete algorithm to load and run on the simulator. The experimental results show that compared with the previous algorithms, the new algorithm inherits the advantages of ensuring the priority of critical tasks, and improves the disadvantages of improper priority selection and late processing of redundant tasks. In the process of task scheduling, the parallelism of the processor is improved, the total task scheduling length is reduced, and the expected effect of the improved algorithm is achieved, which has certain practical application value.
【学位授予单位】：哈尔滨工程大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP332

【引证文献】