近年來，由於大幅地降低了部署和維護網路應用的門限，雲端計算在世界各地已越來越受歡迎。雲端計算的基礎建設因為太龐大以致於能量消耗也相當可觀。另一方面，由於溫室氣體排放和氣候變化的顧慮，能量效率已成為全球最風行的議題之一。在本篇論文中，我們研究對於雲端計算的功率節能機制。我們針對雲端計算中心提出了一個基於緩衝器佔用之雙門限磁滯功率節能機制。為了模仿實際的流量，我們假設連結抵達程序為二狀態MMPP (2-state MMPP)。我們將感興趣的系統塑模成一個三維馬可夫鏈。為了找到穩態機率分佈和其效能指標，我們提出了一種疊代演算法。我們感興趣的效能指標包含系統的平均人數、阻塞機率、立即服務機率、系統延遲、功率節能、總成本。我們研究不同的系統參數對於效能指標的影響，亦即MMPP 轉移機率、連結抵達速率、上下門限、開啟伺服器的數目、功率節能對於阻塞機率的相對重要性。經過大量的數值實驗，我們得到以下的觀察。第一、抵達程序的叢發性越高，總成本越大。第二、系統的平均人數、阻塞率、立即服務機率以及系統延遲都不會受到上下門限的影響。第三、開啟伺服器的數目越多，總成本就會越大。第四、功率節能對於阻塞機率的相對重要性越大，總成本也會隨之增加。最後但並非最不重要的，我們使用電腦模擬程式來驗證解析結果的準確性，而電腦模擬是使用C語言來撰寫。在大多數我們研究的情況中，解析結果和模擬結果是相近的。

In recent years, cloud computing has been more and more popular all over the world as it greatly lowers the threshold for deploying and maintaining web applications. Cloud computing infrastructure is so huge that energy consumption is very significant. On the other hand, energy efficiency has become one of the most prevalent global issues due to the concerns on greenhouse gas emissions and climate change. In this thesis, we study the power saving mechanism for cloud computing. We propose a two-threshold hysteresis power saving mechanism based on the buffer occupancy for a cloud computing center. To emulate the realistic traffic, a 2-state MMPP call arrival process is assumed. The system of interest is modeled as a 3-dimensional Markov chain. An iterative algorithm is proposed to find the steady state probability distribution and the performance measures. The performance measures of interest are the mean number of customers in the system, blocking probability, probability of immediate service, the system delay, the power saving, and the total cost. We study the effect of different system parameters, i.e., MMPP transition rate, call arrival rate, upper and lower thresholds, number of active servers, and relative importance of power saving to blocking, on the performance measures. After extensive numerical experiments, we make the following observations. First, the more bursty the arrival process is, the larger the total cost. Second, the upper and lower thresholds do not affect the mean number of customers in system, the blocking probability, the probability of immediate service, and the system delay. Third, the larger the average number of active servers is, the larger the total cost. The greater the relative importance of power saving to blocking is, the larger the total cost. Last but not least, we verify the accuracy of the analytical results by the simulation program written in C. In most cases studied, the analytical results are close to the simulation results.

[1] B. Furht, "Cloud Computing Fundamentals," Handbook of Cloud Computing, pp. 3-19, Springer, 2010.
[2] Amazon Elastic Compute Cloud, User Guide, API Version ed., Amazon Web Service LCC or Its Affiliate, http://aws.amazon.com/documentation/ ec2, Aug, 2010.
[3] H. W. Ferng, and J. F. Chang, "Connection-wise End-to-End Performance Analysis of Queueing Networks with MMPP Inputs," Performance Evaluation, vol. 43, 2003.
[4] Z. Yang, M. H. Chen, Z. Niu, and D. Huang, "An Optimal Hysteretic Control Policy for Energy Saving in Cloud Computing," GLOBECOM 2011, pp, 1-5.
[5] L. Kleinrock, Queueing Systems: Theory, vol. 1, Wiley-Interscience, 1975.
[6] V. Goswami, S. S. Patra, and G. B. Mund, "Performance Analysis of Cloud with Queue-Dependent Virtual Machines," IEEE Recent Advances in Information Technology, 2012, pp, 357-362.
[7] H. P. Chen, and S. C. Li, "A Queueing-Based Model for Performance Management on Cloud," 2010 6th International Conference on Advanced Information Management and Service (IMS), 2010, pp. 83-88.
[8] H. Khazaei, J. Misi c, and V. B. Mi si c, "Modelling of Cloud Computing Centers Using M/G/m Queues," The First International Workshop on Data Center Performance, Minneapolis, MN, Mar. 2011, pp. 87-92.
[9] H. Khazaei, J. Misic, and V. B. Misic, "Performance Analysis of Cloud Computing Centers Using M/G/m/m+r Queuing Systems," IEEE Transactions on Parallel and Distributed Computing, 2011, pp, 936-943.
[10] H. Khazaei, J. Misi c, and V. B. Misic, "Performance Analysis of Cloud Centers under Burst Arrivals and Total Rejection Policy," IEEE Transactions on Parallel and Distributed Computing, 2011, pp, 1-6.
[11] B. Ciciani, A. Santoro, and P. Romano, "Approximate Analytical Models for Networked Servers Subject to MMPP Arrival Process," IEEE International Symposium on Computing and Applications, 2007, pp, 25-32.
[12] B. Anjum, "MMPP-2 Approximation of Aggregate IPPs with Application to Expedited Forwarding Class in IP-based Networks," ICCT 2011, pp, 494-498.
[13] E. H. Ong, K. Mahata, and J. Y. Khan, "Energy Efficient Architecture for Green Handsets in Next Generatio IP-based Wireless Networks," ICC 2011, 2011, pp, 1-6.
[14] G. Anastasi and L. Lenzini, "QoS Provided by the IEEE 802.11 Wireless LAN to Advanced Data Applications: a Simulation Analysis," Wireless Networks, Vol. 6, Issue 2, pp. 99-108, 2000.