透過您的圖書館登入
IP:3.17.6.75
  • 學位論文

以虛擬平台進行快速且非侵入式地解析應用程式效能

Fast and Non-intrusive Profiling of Application Performance with a Virtual Platform

指導教授 : 洪士灝

摘要


本論文主要是在虛擬平台上建立一個軟體剖析(profiling)機制來協助在嵌 入式系統上的軟體開發工作,此機制主要是建立在QEMU 這個開放原始碼的模擬器 上。QEMU 模擬器廣泛被使用在開發嵌入式軟體的計畫中,如包含在Android 軟體 開發工具組中用來作為Android 開放原始碼平台中之虛擬平台,用以加速並協助 Android 系統上的軟體開發。然而,在現有的QEMU 設計中並沒有估量軟體執行時 間的相關機制,因此我們嘗試在該虛擬平台上建立一套效能分析的架構,我們的 目標是提供快速並且準確的時間評估模型來幫助程式設計師在系統開發的早期估 測程式效能。我們的實驗展示數個不同的時間模型,提供較高精準度的模型需要 較長的模擬時間,使用者可以跟據自己的需求,在模擬時間與準確率之間折衷選 擇。

並列摘要


In this paper, we present a performance analysis framework to support application development on heterogeneous multicore systems with virtual platforms. Our work is based on the open source QEMU simulator which has been widely used to develop software on embedded systems and was included in the Android Software Development Kit. However, to support the software development on multicore platfroms, timing information is critical and thus we attempt to build virtual platforms with the capacity of performance evaluation. Our goal is to provide a framework that gives quick, accurate timing estimation to help multicore application development during early design phases. In this paper, we discuss the construction of various timing models for our virtual platform to simulate program execution.

參考文獻


[2] T. Austin, E. Larson, and D. Ernst. SimpleScalar: an infrastructure for computer
I. Pratt, and A. War eld. Xen and the art of virtualization. SIGOPS Oper. Syst.
the annual conference on USENIX Annual Technical Conference, pages 41 41,
[6] M. V. Biesbrouck, L. Eeckhout, and B. Calder. E cient sampling startup for
sampled processor simulation. In High Performance Embedded Architectures and

延伸閱讀