發(fā)布時間：2019-05-28 02:01

【摘要】：當前,大規(guī)模存儲系統(tǒng)普遍采用副本、糾刪碼等技術(shù)來提供高可靠性。隨著存儲系統(tǒng)規(guī)模和復(fù)雜性的不斷增長,傳統(tǒng)的冗余機制難以提供足夠的可靠性,構(gòu)建高可靠性的存儲系統(tǒng)成為了巨大的挑戰(zhàn)。目前,幾乎所有的硬盤都支持SMART故障預(yù)測技術(shù)。SMART全稱為" Self-Monitoring, Analysis and Reporting Technology",它會監(jiān)控硬盤內(nèi)部的一些與健康狀況相關(guān)的屬性,如果某個屬性值超過了閾值,它就會發(fā)出故障預(yù)警信息。然而,SMART技術(shù)的預(yù)測準確率非常有限,僅能在0.1%的誤報率的情況下預(yù)測出3-10%的硬盤故障。一些研究者研究了基于硬盤的SMART信息,采用統(tǒng)計學(xué)和機器學(xué)習的方法建立硬盤故障預(yù)測模型。然而,這些模型僅能夠在保持較低誤報率的情況下,預(yù)測出60%左右的硬盤故障。 本文嘗試了改進和優(yōu)化基于支持向量機(Support Vector Machine, S VM)的硬盤故障預(yù)測模型。本文還提出了基于反向傳播(Backpropagation, BP)算法的人工神經(jīng)網(wǎng)絡(luò)(Artificial Neural Network, ANN)來建立預(yù)測模型,并且對該BP-ANN預(yù)測模型用AdaBoost算法進行了優(yōu)化。實驗數(shù)據(jù)集來自實際的數(shù)據(jù)中心,包含了多達23,395塊硬盤的SMART記錄。本文采用了新的樣本處理、選取方法以及特征構(gòu)造的方法,提高了模型的預(yù)測準確率。本文還提出了一種基于投票的故障檢測算法,該方法能夠有效降低模型的故障誤報率。為了更加準確地描述硬盤的健康程度(即故障概率),本文也嘗試了使用BP算法建立硬盤的健康度模型。實驗結(jié)果顯示,本文提出的模型取得了非常好的故障預(yù)測性能。優(yōu)化的SVM模型能夠取得最低的誤報率(0.03%),而BP-ANN模型則能在保持較低誤報率時達到95%以上的故障檢測率。本文利用馬爾可夫模型計算了不同存儲系統(tǒng)的可靠性,結(jié)果表明硬盤故障預(yù)測模型可以顯著提高存儲系統(tǒng)的可靠性。此外,本文還初步探索了主動容錯機制在實際大規(guī)模存儲系統(tǒng)中應(yīng)用的問題。
[Abstract]:At present, large-scale storage systems generally use replica, erasure code and other technologies to provide high reliability. With the increasing scale and complexity of storage system, the traditional redundancy mechanism is difficult to provide enough reliability, so it has become a great challenge to build a high reliability storage system. At present, almost all hard drives support SMART fault prediction technology. smart, fully known as "Self-Monitoring, Analysis and Reporting Technology," monitors some of the health-related properties inside the hard disk if a property value exceeds the threshold. It sends out a fault warning message. However, the prediction accuracy of SMART technology is very limited, and only 3 鈮，

本文編號：2486636