基于性能计数器的多核处理器功耗估算-计算机系统结构专业论文.docxVIP

上海交 上海交通大学硕士学位论文 II II ESTIMATING POWER CONSUMPTION OF MULTICORE PROCESSORS USING PERFORMANCE COUNTERS ABSTRACT Multicore processor technology is being more widely used in processor architectures, and power density continues to increase accordingly. thermal management strategy should be introduced into multicore systems. Current processors mainly use dynamic voltage and frequency scaling to alleviate the power and temperature issues. However, From the view of computer architecture, an OS level scheduling strategy can reduce power consumption more effectively. In multicore processor systems, each core of multicore processor has independent computing resource. It is possible to migrate a hot task on an overheated core to other cooler cores to balance power density, and this strategy needs the power of tasks running on each core. This thesis demonstrate a way of power estimation on Core Architecture using performance counters, choosing proper hardware events which are mostly related to CPU power, suggesting a linear model to represent relationship between processors power and hardware events using performance counters, and giving empirical results of the model. The experimental result shows that this power model can estimate CPU power of different tasks; operating systems can schedule tasks between hot cores and cool cores based on the estimated power of the tasks. This work provides the model parameters required by an energy-aware real time online scheduler. Key Words：multicore processor, performance counter, power model, parametric estimating III III 目 录 第一章 绪论1 1.1 研究背景 1 1.2 国内外研究现状 3 1.3 本文的组织结构 6 第二章 操作系统的任务调度8 2.1 进程的调度策略 8 2.1.1 先来先服务算法9 2.1.2 时间片轮转调度算法9 2.1.3 最短进程优先算法9 2.1.4 最短剩余时间算法10 2.1.5 反馈队列算法10 Linux 进程调度 11 Linux 进程分类 11 Linux 进程的状态 12 Linux 调度器 13 2.3 当前调度策略的不足 14 第三章 基于温度与功耗的调度思想 16 3.1 基本思想 16 3.2 基于传感器的方法 16 3.2.1 方法原理16 3.2.2 从内核中读取温度17 3.3 基于性能计数器的方法 18 3.3.1 硬件性能计数器18 3.3.2 功耗估算19 3.4 基于温度和功耗的任务调度 20 3.4.1 根据功耗预测温度20 3.4.2 基于温度与功耗的调度策略2