程序層級上耗電預估參數與公式的校正

Abstract

搭載電池的移動式裝置經常受到能源上嚴格的限制。程序階級上的耗能分析工具可以找出系統中最耗能源的程序，並且可以仔細地分析出個別硬體元件的耗能情形。利用這樣的工具，軟體開發工程師可以分析與微調各程序的電能消耗，藉此來提高電池的使用時間。不過這種耗能分析工具經常綁定特定的硬體，所以需要為各個硬體平台來做耗能分析工具的校正。此外，對於新加入的硬體元件也需為其創造新的耗能預估公式。這篇論文提出一個兩階段式的方法來校正產品上的耗能分析工具。第一階段利用數位電表來重新建立屬於新產品的耗能功率表。第二階段則是使用線性回歸分析來創造新的耗能預估公式。在五種情境下驗證的結果顯示，經過我們校正之後耗能分析工具的預測錯誤率都低於10%。此外，我們發現FTP上傳與下載的程序雖然消耗的電能不同，但是花在CPU計算與網路傳輸的耗能比例卻是一樣的。

The battery-powered mobile devices get tight constrains on energy resources. The process-level energy profiling tools can identify the most energy-consuming process and detail the energy usages of each hardware component. With the help of energy profiling tools, programmers can fine-tune the energy consumption of processes to improve the battery lifetime. However, the profiling tools are highly hardware dependent and therefore require to be calibrated for each hardware platform. Besides, new energy estimation formulas need to be created for new hardware components. In this thesis, a two-phase calibrating approach is proposed to handle the two issues on off-the-shelf product devices. The first phase reconstructs the power table with a power meter, and the second phase creates new energy estimation formulas with the linear regression analysis. The accuracy of the calibrated tool is evaluated under five scenarios with the error ratios proven below 10%. Moreover, the energy consumption of FTP upload and download processes is different but the ratio of CPU computing energy to networking energy is the same.