適用於嵌入式處理器之高階軟體耗能評估工具

Abstract

為了處理日趨複雜的嵌入式軟體並盡量降低嵌入式處理器的耗能，近年來，提昇能源效率 (energy-efficiency) 已經成為設計嵌入式系統時的一大重點，在提昇能源效率的研究當中，軟體耗能評估工具經常被用來評估一項電源管理政策的優劣以及它對嵌入式軟體之執行時間與耗能的影響。雖然目前已經存在許多套軟體耗能評估工具，但它們大多需要花費很多時間去評估耗電，或是沒有考慮到處理器的電源管理狀況，因此並不適合在較複雜的嵌入式系統當中被用來評估支援電源管理 (power management) 功能之處理器的耗電。本論文提出之高階軟體耗能評估工具 SEProf 即是一套能夠在具作業系統之嵌入式系統裡，快速評估支援電源管理功能之處理器耗電的工具，我們已經將 SEProf 實做於 Linux 核心 2.6.19 當中，並以一個具 ARM11 MPCore 處理器的嵌入式系統平台為例做了許多實驗，實驗結果顯示在大多數的情況之下，SEProf 所評估出來的平均功率誤差是在 2% 以內，而功率誤差的標準差則是在 6% 以內。

In order to handle the growing complexity of embedded system while minimizing its energy consumption, energy-efficiency of embedded software recently has become one of the most important issues in the design of embedded systems. To examine the energy-efficiency of embedded software, estimation of the energy consumption of embedded software is very critical. Although a number of software energy estimation tools have been proposed, most of them are not able to efficiently estimate the energy consumption of embedded processors with power management features, e.g. dynamic voltage and frequency scaling (DVFS). Therefore, this paper presents a high-level software energy profiling tool, called SEProf, targeted toward energy estimation of embedded processors on operating system (OS)-based embedded systems with power management functions enabled. We implemented the proposed SEProf in Linux kernel 2.6.19 and conducted a number of experiments on an ARM11 MPCore processor. The experimental results demonstrate that the average error of the power estimation using SEProf is within 2%, and the standard deviation of the power estimation error is under 6% in most cases.