【作者】 黄江伟；

【导师】 陈天洲；

【作者基本信息】 浙江大学 ， 计算机科学与技术， 2009， 博士

【摘要】 半导体技术的快速发展,为嵌入式处理器技术的不断提升创造了条件。随着处理器的集成度、复杂度的增长,以及性能的不断提高,其功耗也节节攀升。近年来,处理器单位面积的功耗已呈指数级增长。功耗的迅速增长已经成为制约处理器进一步发展的一个主要因素。而大多数的移动嵌入式系统,由于其应用环境的特殊性,不得不采用电池供电。相比于有线电源,电池的容量有限,因此移动嵌入式系统对功耗具有更加严格的要求。但相对于处理器技术的快速发展,电池技术发展严重滞后,因此如何降低嵌入式系统本身的功耗,合理延长现有电池供电时间已成为研究者关注的主要热点。嵌入式系统是一个软硬件混合体,其中硬件的运行直接导致能量的消耗。而硬件的行为都是通过软件进行控制的,因此硬件层之上的软件也对电能的消耗起着举足轻重的作用。通过软件对硬件行为进行控制,可有效降低系统功耗。因此涌现了一批硬件节能降耗手段,并提供相应的软件接口,尤其是以DVS技术为代表的处理器节能技术,为嵌入式软件提供了控制硬件功耗的有效机制。本文主要以DVS技术为基础,在以电池驱动的移动嵌入式系统中,从系统软件层面研究节能计算,以达到有效利用硬件节能降耗技术进行系统节能优化。特别是在保障系统实时性的同时降低系统能耗,从而获得电池放电期内性能最大化。本文主要从以下三个方面进行了研究:1、本文首先分析了移动嵌入式系统中的电池模型,以此为出发点研究了DVS技术支持的低功耗设计问题。DVS技术在调频调压过程中会影响到负载电流强度,从而进一步影响电池的放电量,因此需要研究电池模型在DVS技术下的修正。本文通过研究电池放电量和DVS的关系,建立了一种电池模型驱动的低功耗嵌入式系统的设计模型。通过在设计过程中引入电池特性、量化DVS对电池容量的影响,建立了电池模型驱动的低功耗嵌入式系统模型,并根据该模型推导出了低功耗嵌入式系统的设计指导原则。2、在上述硬件平台设计方法的支持下,本文还系统地研究了操作系统对低功耗调度的支持,特别是在实时调度和存储设备访问请求调度情况下的低功耗调度方法。本文首先研究了实时任务模型在DVS下的工作机理,通过离线计算,在保证任务集实时性的前提下,尽量在电池供电期内完成足够多的任务,形成了Offline任务集低功耗运行速度决策模型。由于离线计算是静态的,而操作系统处于运行时,能够获得整个系统的运行时信息,因此本文还研究了在线时的低功耗调度方法,完成了Online任务集低功耗调度决策模型。此外,本文针对操作系统对存储层次的访问进行低功耗优化,设计了硬盘设备功耗管理和访问请求调度,优化了硬盘设备的访问请求,降低了硬盘功耗。3、本文同时利用编译器对执行程序内部特性的了解,研究了通过编译器对执行程序进行改造,在运行时为操作系统提供低功耗调度支持。在此部分内容中,本文首先研究了如何利用编译器对程序进行细粒度、自适应的DVS节能改造。然后,通过编译器分析执行程序存储行为特性,建立了程序与任务的行为和功耗模型。在此基础上,设计了操作系统和编译器协同方式,为操作系统低功耗调度提供细粒度的支持。最后,本文分别针对基于电池模型的软硬件协同设计,实时系统低功耗调度以及编译器对操作系统的支持进行了实验验证和分析。实验结果表明本文所描述的基于DVS技术的移动嵌入式系统软件节能的研究,可分别从系统软件的不同层面降低系统功耗。本研究充分利用了DVS技术的优势以及电池特性,从电池与处理器等硬件层面出发,研究了DVS技术对低功耗实时调度、节能编译器等方面的影响,系统的给出了相应的解决方案。更多还原

【Abstract】 With rapid development of semiconductor technologies, embedded circuit technology has been improved constantly. The integration and complexity of the chip is increased, as well the size is getting smaller and smaller. As the performance of processor increased, the power consumption of processor is bigger than ever before. Because of the improved integration and complexity, the power consumption is rising. Power consumption per unit area is rising linearly as circuit area increase. The rising of power consumption per unit becomes a constraint for processor development. Most of the mobile embedded systems are driven by battery. Because of the limitation of battery capacity, mobile embedded systems should be more power aware. Compared to the rapid development of processor technology, battery technology is lag behind. So extending the limited battery life is one of the important topics for portable embedded systems.Embedded system is a mixture of hardware and software. Power is consumed by hardware running. Software is the controller of hardware, and it drives the hardware to run. So software is the sole of embedded system, and it is consuming power by driving the hardware run indirectly. For power aware computing, the behavior and characteristic of software should be taking into account. Especially, the DVS technology provides an effective way for software to reduce the hardware power consumption. By the help of DVS, software can adjust the core supply voltage of processor dynamically.In this thesis, it focuses on applying DVS to mobile embedded system. There are some questions about using DVS in mobile embedded system. One question is what about the battery behavior will be affected by DVS, and the other is what about the effective of using DVS in real-time embedded system. So this thesis makes the following contributions:(1)First, it presents a battery driven power aware design method for mobile embedded system with DVS support. The discharged capacity of battery is not a constant, which will be affected by current intensity. And the current of battery is decided by hardware running status. So in this thesis, the relationship between battery capacity discharged and DVS is studied. Then this thesis gives a power aware mobile embedded system design method driven by battery. This method quantifies the DVS impact on battery capacity, and gives a theory model about power aware design. According to this model, it deduces a principle for mobile embedded system power aware design.(2)It proposes a power aware operating system scheduling and a power aware dispatching for storage devices access requests, especially for real-time system. It first studies the real-time task model with DVS support, and use offline schedule to try completing the tasks as much as possible during the battery life with the task deadline guaranteed. As the offline schedule is static, and it can’t gain any running information. Operating system could collect the running information for whole system or any tasks. On this basis, an online schedule algorithm is developed to enhance the offline scheduler. Power consumption for peripheral equipments is also very important. This thesis takes HDD as an example, to study the power reduction technology for HDD using access requests schedule.(3) This thesis also studies how to use compiler to support lower power scheduling. The compiler is used to analyze the memory behavior of task, and build a task power model. Using this model, compiler gives a static DVS decision for task. And it also studies the cooperative protocol between Operating system and Compiler. Operating system uses the fine-grained information from compiler static decision in power aware task scheduling. This fine grained information would give some hint for online schedule.At last, this thesis evaluates the proposed power aware method and optimization. Experimental results show that these optimization methods can effectively reduce the power consumption for whole system. This research fully utilizes the characteristics of DVS technology and battery. It studies DVS technology effective for low power real-time scheduling and compilers from battery model, and gives a system level solution.更多还原