线性加速比并行实时任务的节能算法研究

发布时间：2018-08-04 21:35

【摘要】：嵌入式设备的性能大幅度提高,对能源的需求不断上升。能耗的增长不仅制约了嵌入式系统集成度的增加,并且需要复杂的散热技术,这更是提高了产品成本。因此,怎样更有效的节能已成了嵌入式系统研究的热点问题。另一方面,随着多核处理器技术在服务器和PC机上的广泛使用,多核技术表现出同一频率下,性能提高快,功耗密度低等特点,使得多核成为嵌入式系统未来的统治平台。现有的多核节能调度方面的研究主要是针对串行任务(sequential tasks)模型,该模型假设一个任务在同一时刻只能运行在一个核上,而不能同时运行在多个核上,这无法发挥出多核技术在性能上的优势。因此,本文针对并行任务(parallel tasks)模型,即任务可以在处理器所有核上同时运行的节能问题进行了研究。实际系统中处理器功耗是一组和频率一一对应的离散值。现有的实时系统能耗的讨论为了简化模型,往往将处理器频率值假设为连续改变的,然而这并不适用于实际系统。因此,本文研究更接近真实处理器频率改变情况的离散处理器频率模型中,线性加速比模型下的多核并行实时任务的节能调度问题,提出了两个满足时间约束,采用Gang scheduling策略的并行任务能耗优化算法。本文假设所有处理器核都支持DVS技术且加速比是线性的,在此条件下,本文首先给出一个定理证明了当各任务执行在系统全部核上时系统能耗最小。然后,将问题建模为一个0-1整数线性规划,利用最早截止期优先算法确定任务调度,并给出了两个高效的节能算法确定各任务的工作频率。论文最后通过大量的模拟实验,证明本文提出的算法的效果显著,可以达到几乎与最优解相同的节能效果。
[Abstract]:The performance of embedded devices has been greatly improved, and the demand for energy has been rising. The increase of energy consumption not only restricts the increase of embedded system integration, but also requires complex heat dissipation technology, which increases the cost of products. Therefore, how to save energy more effectively has become a hot issue in embedded system research. On the other hand, with the wide use of multi-core processor technology in server and PC, multi-core technology has the characteristics of fast performance improvement and low power density at the same frequency, which makes multi-core system the dominant platform for embedded systems in the future. The existing research on multi-core energy-saving scheduling is mainly focused on the serial task (sequential tasks) model, which assumes that a task can only run on one core at any one time, but not on more than one core at the same time. This does not give play to the performance advantages of multiple nuclear technologies. Therefore, this paper focuses on the parallel task (parallel tasks) model, that is, the energy saving problem in which the task can run on all the cores of the processor at the same time. Processor power consumption in a real system is a set of discrete values corresponding to frequency. In order to simplify the model, the current discussion on the energy consumption of real-time systems often assumes that the processor frequency values are continuously changed, but this is not suitable for real systems. Therefore, in the discrete processor frequency model, which is closer to the real processor frequency change, the energy saving scheduling problem of multi-core parallel real-time tasks under linear speedup model is studied, and two satisfying time constraints are proposed. Parallel task energy consumption optimization algorithm based on Gang scheduling strategy. In this paper, we assume that all processor cores support DVS technology and the speedup is linear. Under this condition, we first give a theorem to prove that when each task is executed on all cores of the system, the energy consumption of the system is minimum. Then, the problem is modeled as a 0-1 integer linear programming, the earliest deadline first algorithm is used to determine the task scheduling, and two efficient energy-saving algorithms are given to determine the working frequency of each task. At last, through a lot of simulation experiments, it is proved that the proposed algorithm has a remarkable effect and can achieve almost the same energy saving effect as the optimal solution.
【学位授予单位】：东北大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP368.1

【相似文献】