適用於模擬的多核心平台中的效能監測與預估

Performance Monitoring and Estimation on Emulated Multicore System

林庚佑(Geng-You Lin)

指導教授 ： 洪士灝

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2010年)

https://doi.org/10.6342/NTU.2010.10470

摘要

在硬體開發期間，如何讓軟體開發者可以在虛擬平台上快速與精確地模擬多核心系統，是當前很重要的研究議題。
本論文敘述我們在虛擬平台中開發出一個適用於效能監測與預估的機制，提供多個時間評估模型，讓使用者以虛擬時間裝置（Virtual Timing Device）選擇所需的模擬速度與精確度。
使用者可以花較多時間在程式重點部分，做教精確的模擬，而不需要花太多時間去模擬其他較不重要的部分。
我們也針對模擬速度的部份，提出改進的機制來加速整個的模擬速度，並且可以搭配我們實驗室另一個智慧型程式追蹤工具(Smart Event Tracing)，讓使用者能容易地在虛擬平台做程式分析。
最後，我們也針對模擬速度的部分做了最佳化的技術。
而實驗結果也證明了我們的虛擬平台的執行速度也足以支援複雜的作業系統以及多核心程式而且我們的時間評估模型與效能計數器(performance counter)也提供了夠精確的效能資訊。

關鍵字

虛擬平台 ； 嵌入式系統 ； 效能分析

並列摘要

Before the hardware of a system is available, developers often use a virtual platform to develop software for the system. However, to deliver a fast and accurate simulation of multicore systems is still an important issue that is open for research. In this paper, we describe an integrated framework for performance monitoring and estimation on emulated multicore systems.
First, we discuss the timing models implemented in the virtual timing device, which can be adjusted dynamically in our virtual platform to estimate the execution time of a program with di erent speeds and degrees of accuracy, so developers can focus on important parts of a program, instead of wasting time to simulate important parts. Then, our virtual platform supports the Smart Event Tracing tool developed in our research group for system developers to pro le and analyze the performance of a system.
Finally, we proposed several optimization techniques to improve the speed of our virtual platform. Our experimental results showed that our virtual platform is fast enough to support complex systems in running operating systems and multicore applications, and our timing models and performance counters provided performance data with good accuracy.

並列關鍵字

virtual platform ； embedded systems ； performance analysis

參考文獻

[2] Doug Burger and Todd M. Austin. The SimpleScalar tool set, version 2.0. SIGARCH Comput. Archit. News, 25(3):13{25, 1997.

[4] Gilberto Contreras and Margaret Martonosi. Power prediction for intel XScale®

[6] Kemal Ebcioglu, Erik Altman, Michael Gschwind, and Sumedh Sathaye. Dynamic binary translation and optimization. IEEE Trans. Comput., 50(6):529{ 548, 2001.

2255{2260. IEEE Computer Society, 2010.

Marina Del Rey, California, 2002. ACM.