雲端智慧型效能評估

Intelligent Performance Evaluation in Cloud Computing

林胡揚(Yang Lin Hu)

指導教授 ： 王凡

國立臺灣大學/電機資訊學院/電子工程學研究所/碩士(2011年)

https://doi.org/10.6342/NTU.2011.10225

摘要

雲端運算已是一股勢不可擋的趨勢，隨著雲端運算的發展，軟體亦漸漸從本地端移植至網路服務。傳統上，軟體效能僅取決於執行軟體的硬體運算效能，是可被預測的；而網路服務有別於傳統，是被小由數個虛擬主機、大至數百個數據中心所組成的雲端設施所執行，對於服務開發者與使用者來說，共享資源的雲端是個巨大的黑盒子，只能藉由真實發送存取需求來量測效能，且難以預測。在本論文中，我們提出利用支向機進行機器學習的架構作為網路服務效能評估的手段，使用監督式學習演算法的特性來捕捉在雲端環境中的效能不穩定因素以及真實使用者在背景存取服務行為對於網路服務效能的影響；我們亦提出一種特徵選擇方法來降低取樣資料數，並提升評估的準確度。實驗結果顯示，我們所提出的架構確實可以持續地對於雲端環境中的網路服務進行效能預測評估，且達到高準確度。

關鍵字

雲端運算 ； 效能評估 ； 效能預測 ； 網路服務 ； 機器學習 ； 支向機

並列摘要

Cloud Computing is a trend can't be halted. More and more applications or software had been porting as Web service in Cloud. Traditionally, software performance is determined by the host hardware throughput, and is predictable. Web service is differs from traditional software, Web service is hosted by Cloud infrastructure that might be assembled from several virtual machines to hundreds of datacenters. Sharing resources Cloud is a big black-box to developers and users. We can only measure performance by actually sending requests, and the performance of Web service is hard to predict. In this thesis, we propose a framework integrated with Support Vector Machine (SVM) as a mechanism to evaluate performance of Web service. We use the characteristics of supervised learning algorithm (SLA) to capture the elements of unstable performance in Cloud, and the effects of real behavior of background usage. We also propose a feature selection scheme to reduce size of sampling data. Our experimental results show that the framework we proposed can evaluate/predict performance continuously, and achieve high-accuracy.

並列關鍵字

Cloud Computing ； Performance Evaluation ； Performance Prediction ； Web service ； Machine Learning ； Support Vector Machine

參考文獻

[4] Zhichun Li, Ming Zhang, Zhaosheng Zhu, Yan Chen, Albert G. Greenberg, and Yi-

[14] Tom M. (Tom Michael) Mitchell, Machine Learning, McGraw Hill, Oct 1997.

[1] S. Schach, Software Engineering, McGraw Hill, 1992.

[2] Chih-Chung Chang and Chih-Jen Lin, “LIBSVM: A library for support vector

27:1–27:27, 2011, Software available at http://www.csie.ntu.edu.tw/