雲端數據機房散熱系統最佳化演算法

Abstract

雲端運算(Cloud Computing)為近年科技大廠紛紛跨足的產業，雲端應用必須依靠穩健而有力的機房來負擔運作，因此數據機房內的系統軟體和硬體架構是協助完成客戶需求的重要關鍵。為了負荷龐大的運算和儲存設備的處理效能，數據機房必須不中斷地提供服務，因而容易出現熱當機的情形，故散熱問題成為機房管理者所面臨最大的難題。散熱問題通常與耗電量、耗水量和碳排放量有關，故可視為一個多品質問題。目前中外文獻都是以工程角度提出改善散熱系統方法，尚未見有文獻利用統計方法提出解決之道。因此，本研究之主要目的為利用實驗設計(Design of Experiments, DOE)發展一套多品質同時最佳化演算法解決雲端數據機房散熱問題。本論文先利用實驗設計法規劃實驗，再結合加權主成份分析(Weighted Principal Component Analysis, WPCA)及反應曲面法(Response Surface Method, RSM)提出一套多品質特性同時最佳化之演算法。本研究最後以一個模擬之雲端數據機房散熱系統案例來說明本論文所提出之多品質最佳化演算法確實有效可行。

Cloud computing is an emerging industry and is viewed as a new business model of IT service in recent years. Cloud computing must rely on robust and powerful data center to fulfill the need of customers. The data center must keep running in order to maintain the high efficiency of computing and storage service. This situation might cause computer crashed due to overheat. Therefore, the heat dissipation problem is a serious challenge for the supervisor of data center. The heat dissipation problem is usually associated with power consumption, water consumption and carbon emissions, and can therefore be considered as a multi-response problem. Previous studies related to heat dissipation problem are based on engineering method to improve the cooling system, and it is rarely seen that statistical methods are utilized to solve the heat dissipation problems. Therefore, the main objective of this study is to develop a multi-response optimization algorithm using Design of Experiment (DOE), Weighted Principal Component Analysis (WPCA) and Response Surface Method (RSM) to find the optimal parameter-setting of cooling system for data center of cloud computing. Finally, a simulated case is used to demonstrate the effectiveness of the proposed procedure.