Android系統上,開機、網頁瀏覽、串流播放的執行時間剖析和瓶頸分析

Abstract

Android系統在開機、網頁瀏覽及串流播放這三種使用情境下的效能表現顯得不盡理想。在Android系統上剖析一種使用情境的執行時間，會發現面臨到三個主要的特性:第一、一個使用情境的執行流程會跨不同軟體層呼叫不同元件，第二、任一軟體層都是由一種以上的程式語言所構成，第三、系統有限的儲存空間。本論文提出一個可以在多語言多軟體層平台上找出執行時間瓶頸的分階段反覆插入剖析方法。由單一的模組開始，進而逐漸地加入不同軟體層的不同模組來剖析，以避免在剖析過程中產生大量不必要的資料，最後再把不同軟體層執行時所輸出的資料合併分析，藉此找出執行時間瓶頸。此方法被實做測試在一台Android的產品上，實驗結果顯示72%的開機時間花在user space環境的初始化，而在user space初始化中，44.4%的時間花在啟用背景服務程序及背景管理程序，37%花在先行載入Java classes和resources.實驗結果並顯示，網路是影響網頁瀏覽最主要的因素。在載入一個2128 kB大小的網頁的實驗環境下，螢幕繪圖顯示系統僅佔總執行時間的5%。在Wi-Fi環境下播放一段22 MB的串流短片，系統需要花5.7%的時間來做撥放準備的動作。影片資料下載加上資料解碼部分的執行時間共佔了這段準備時間的72%。

Android-based systems perform slowly in three perceptible scenarios: booting, browsing, and streaming. Time profiling on Android devices encounters three unique properties: 1) the execution flow of a scenario invokes multiple software layers, 2) each software layer is implemented in different programming languages, and 3) log space is limited. This thesis proposes a staged iterative instrumentation approach that starts profiling a scenario from a single module, restrainedly profiles more modules and layers to avoid enormous irrelevant profiling results, and finally consolidates the profiling results from different layers to find out the bottlenecks. Experiments on the off-the-shelf Android product showed that 72% of booting time is spent on the initialization of user-space environment; specifically, 44.4% of user-space initialization time is to start Android services and managers, and 39.2% is for preloading Java classes and resources. Experimental results also showed that the networking technology is the most significant factor influencing the browsing performance on Android. The time of drawing screen only takes less than 5% of total time for browsing a 2128 kB web page. In the streaming scenario, video preparation causes 5.7% time overhead for playing a 22-MB video file over Wi-Fi connection. Execution time of Video-downloading and data-decoding take 72% ratio of preparation time.