網路的日益漸進，虛擬化技術已是在雲端儲存中十分實用的技術之一，然而也伴隨著許多的問題，如TASK分配、硬體IO輸出、檔案錯誤等問題。就檔案錯誤部分，為了解決該問題，供應商需不斷對虛擬化系統進行備份的動作，然而因備份需讓虛擬機器關機，故勢必會影響到使用者使用該系統的方便性，為了因應此狀況，提出了在不關機的狀態下進行備份的技術，這就是Snapshot。
然而，Snapshot策略運行時，需不停對該系統檔案狀況做紀錄和複製，這些動作會影響硬體I/O輸出以及VCPU的TASK分配，造成虛擬化系統在讀寫效能上的影響。而如何在進行Snapshot策略時，盡量降低檔案讀寫效能的影響，並用Snapshot正確恢復到指定的檔案系統狀態，則是近期討論的主題之一。然而，許多策略已被提出，如Logic Volume Management(LVM)是針對檔案系統以Physical Extent(PE)為單位進行比對複製，並用此來對目前的檔案系統更動來進行恢復，此方式雖可順利恢復但卻導致無法恢復的狀況發生；Parallax是在檔案寫入時，將更動的資料區塊強行寫入新資料區塊並將舊資料區塊收集來生成Snapshot，恢復機制則同於LVM故仍有相同問題；SNPDisk則是檔案要寫入檔案系統前，利用共享空間將Memory中的檔案複製並生成Snapshot，而進行恢復機制時由原始的Snapshot依序往下進行恢復，恢復時遇到損壞的檔案則該檔案不進行恢復以避免系統發生崩潰。然而這些方法皆無顧慮到，當進行恢復機制時，如發生Snapshot損毀，應如何使檔案恢復到使用者所期待的檔案狀態。
本論文對此提出較佳的解決方法，將Snapshot做增量備份，藉以將檔案恢復程度提昇，盡量使檔案恢復到指定的檔案狀態。所提出的策略主要有幾點：其一，做增量備份時，需盡量減低儲存空間消耗和檔案讀寫效能消耗；其二，進行Snapshot恢復時，如果發生Snapshot損毀時，才會進行修復動作；其三，我們的策略主要是強化SNPDisk所提出的方法，但並不降低其優勢。
我們的策略研究於XEN虛擬化系統上，並會把近年以及提出的策略在XEN中進行評估。在Snapshot生成方面，將使用Postmark來進行讀寫效能損耗評估並依結果證實，我們的生成機制並不會過於消耗系統讀寫效能。而Snapshot恢復方面，將以C語言進行恢復程度模擬和恢復時間模擬並以結果證實，我們的恢復機制比其它的恢復機制在Snapshot損毀時擁有較好的恢復程度，以及我們的Snapshot恢復機制中增加修復條件判斷的必要性。

With the development of the Internet, virtualization technology has become one of the useful cloud storage technologys. However, it is brought with several problems such as TASK allocation, hardware I/O, and file error etc. To solve the file error problems, suppliers need to constantly duplicate the visualization system. Nonetheless, the duplication has to be conducted when the machine is off, which absolutely affect user’s convenience. Under this circumstance, Snapshot technology is proposed, which allows the duplication work while the machine is still on.
	Nevertheless, as Snapshot works, it has to endlessly record and duplicate the system’s file state, which influences hardware I/O and VCPU’s TASK allocation, and causes deficiencies in the visualization system’s read and write performance. Thus, ways to avoid deficiencies in the system’s read and write performance when performing Snapshot method and using Snapshot to recover the file system to the certain state has become the topic for discussion. Many methods are already proposed such as Logic Volume Management (LVM), which is using Physical Extent (PE) as the unit of file matching duplication, and to recover the changes in the file system. LVM does help recover the file system but sometimes it fails. As to Parallax, it compulsively writes in the new data block with the altered data block and then collects the old data block as Snapshot, and the recovery mechanism is alike to LVM so the same problem exists. SNPDisk uses the shared space to replicate files in Memory and generates into Snapshot before files are written in the system. The recovery mechanism is conducted in turn by the original Snapshot, in order to avoid system breakdown, the damaged files won’t be recovered. But, the mentioned methods above all neglect that, when conducting recovery mechanism, what should be done to fit the user’s expectation in case the Snapshot breaks down. 
	This thesis aims to propose a better solution: doing incremental backups for Snapshot to enhance files recovery, and recovering files to specified state to the best. The proposed methods are: first, to minimize the consumption of the storage space and the read and write performance; second, when Snapshot rolls back, the recovery executes only when Snapshot breaks down; third, our method is primarily to strengthen the SNPDisk method without degrading its superiority. 
	Our method is to research XEN visualization system, and will take methods proposed in recent years as evaluating factor within XEN. In the generation of Snapshot, we will use Postmark to evaluate the consumption of read and write performance, and to confirm that our generation mechanism doesn’t over consume the system’s read and write performance in accordance with the experiment result. In the Snapshot recovery, we will conduct recovery level simulation and recovery time simulation with C programming language, and to confirm our recovery mechanism is better than others when the Snapshot breaks down, and the necessity of adding factors of recovery conditions in.

第一章、緒論	1
1.1  論文簡介	1
1.2  論文架構	4
第二章、背景知識	5
2.1  Cloud Service	5
2.2  Virtualization	5
2.2.1  Xen	6
2.3  Error For Data Storage	7
2.3.1  何種情況檔案會發生錯誤	8
2.3.2  Snapshot	9
第三章、相關研究	10
3.1  Snapshot策略	10
3.1.1  Logic Volume Management(LVM)	11
3.1.1.1  Logic Volume Management(LVM) 缺陷[8]	12
3.1.2  Ventana	14
3.1.3  Parallax	16
3.1.4  SNPdisk	20
3.2  問題點描述	23
3.3  考慮點描述	24
3.3.1  檔案讀寫速度影響	24
3.3.2  恢復程度探討	26
3.3.3  修復檔恢復速度探討	27
第四章、我們的新策略	28
4.1  設計原理	28
4.2  設計內容	29
4.3  Snapshot恢復機制描述	36
4.3.1  修復Snapshot檔機制	38
4.3.2  判斷修復機制	40
4.3.3  Snapshot修復探討	43
第五章、實驗及模擬	44
5.1  實驗環境與測試工具	44
5.2  理論分析	45
5.3  模擬結果	47
5.3.1 檔案讀寫效能指標	47
5.3.2  Snapshot修復能力指標	52
5.3.3  Snapshot恢復時間指標	55
5.3.4  Xor評估模擬	61
5.3.5  損壞率不同時的恢復程度趨勢	63
第六章、結論與未來工作	66
第七章、參考文獻	69

圖目錄
圖2.1、Xen半虛擬化基本架構	7
圖3.1、時間對照線	12
圖3.2、Ventana Branches	15
圖3.3、Parallax的Snapshot生成機制	17
圖3.4、Parallax的Snapshot恢復機制	19
圖3.5、SNPDisk的Snapshot結構	20
圖3.6、SNPDisk的Snapshot生成機制	21
圖3.7、SNPDisk的Snapshot恢復機制	23
圖4.1、Snapshot生成機制流程圖比較	35
圖4.2、Snapshot恢復機制流程圖比較	40
圖4.3、判斷機制流程圖	42
圖5.1、時間點對照圖	45
圖5.2、檔案讀寫效能實驗結果圖	47
圖5.3、檔案損壞個數和恢復程度關係圖	52
圖5.4、恢復時間圖(公式模擬)	55
圖5.5、恢復時間圖(實際模擬)	55
圖5.6、恢復時間圖(SNPDisk公式模擬比較)	56
圖5.7、恢復時間圖(Our method(有判斷機制)公式模擬比較)	56
圖5.8、恢復時間圖(Our method(無判斷機制)公式模擬比較)	56
圖5.9、Xor最佳化模擬圖	61
圖5.10、損壞率對恢復程度模擬圖	63

 
表目錄
表2.1 LVM Snapshot內容對照表	13
表4.1 Xor為1時的恢復程度	32
表4.2 Xor對照恢復程度及儲存空間額外消耗表	33
表5.1 檔案讀寫效能實驗結果數據	47
表5.2 恢復程度參數設定	53
表5.3 Snapshot恢復時間參數設定	58
表5.4 Xor次數最佳化參數設定	62
表5.5 Xor對照恢復程度表	62
表5.6 Xor對損壞率運算恢復程度表	64
表5.7 Xor對照損壞率及恢復程度表	65

[1] Citrix Systems Inc., "Xen 3.0 Virtualization User Guide " , http://bits.xensource.com/Xen/docs/user.pdf
[2] Red Hat, Inc., "LVM Architectural Overview" , https://access.redhat.com/site/documentation/en-US/Red_Hat_Enterprise_Linux/6/html/Logical_Volume_Manager_Administration/LVM_definition.html
[3] S. Quinlan and S. Dorward. "Venti: A new approach to archival storage." In FAST '02: Proceedings of the Conference on File and Storage Technologies, pages 89.101, Berkeley, CA, USA, 2002. USENIX Association.
[4] B. Pfaff, T. Garfinkel, and M. Rosenblum, "Virtualization Aware File Systems: Getting Beyond the Limitations of Virtual Disks," Proc. USENIX Symp. Networked Systems Design & Implementation, May 2006, pp. 353–66.
[5] D. Meyer et al., "Parallax: Virtual Disks for Virtual Machines," Proc. ACM SIGOPS/EuroSys Euro. Conf. Comp. Sys. (EuroSys , May 2008, pp. 41–54.
[6] A. Warfield et al., " Parallax: Managing Storage for A Million Machines," Proc. USENIX Hot Topics in Op. Sys., June 2005, pp. 1–11.
[7] Lei Yu; Chuliang Weng; Minglu Li; Yuan Luo ," SNPdisk An Efficient Para-Virtualization Snapshot Mechanism for Virtual Disks in Private Clouds" Network, IEEE Volume: 25 , Issue: 4 , Publication Year: 2011 , Page(s): 20 - 26
[8] W. Xiao and Q. Yang, " Can We Really Recover Data If Storage Subsystem Fails?," Proc. Int’l. Conf. Distrib. Comp. Sys., June 2008, pp. 597–604 .
[9] J. Katcher, "Postmark: A New System Benchmark," tech. rep. 3022, Network Appliance, Inc., Oct. 1997.
[10] "Xen-Users Common Questions Guide",http://www-archive.xenproject.org/files/Support/XenUsersCommonlyAskedQuestions.pdf
[11] K. Fraser,S. Hand,R. Neugebauer,I. Pratt,A. Warfield,and M. Williamson.,"Safe Hardware Access with the Xen Virtual Machine Monitor",1st Workshop on Operating System and Architectural Support for the on demand IT InfraStructure (OASIS), Oct 2004.
[12] Red Hat, Inc.,"LVM Administrator's Guide",https://www.centos.org/docs/5/html/Cluster_Logical_Volume_Manager/index.html
[13] J.-F. Chiu and W.-H. Hao, "Mutual-Aid: Diskless Checkpointing Scheme for Tolerating Double Faults", Proc. IEEE Symp. High Performance Computing and Communications ( HPCC'08), Sep. 2008, pp.540-547.
[14] W. Xiao,J. Ren,and Q. Yang,"A Case for Continuous Data Protection at Block Level in Disk Array Storages,"IEEE Trans. Parallel and Distrib. Sys.,vol. 20, no. 6, Jun. 2009, pp. 898–911.
[15] Z. Peterson and R. Burns."Ext3cow: a time-shifting file system for regulatory compliance."ACM Transactions on Storage, 1(2):190–212, 2005.
[16] ZHENG Tingting. WU Yanjun. HE Yeping.,"Rapid virtual machine cloning in cloud computing environment.",Computer Engineering and Applications,2011 47（13）:63-67.
[17] Chen B, Xiao N, Cai ZP, Wang ZY. "Prefetch mechanism for on-demand software deployment in virtual machine environments." Journal of Software, 2010,21(12):3186−3198.
[18] 褚祸勇, 肖 侬, 蔡志平, 等."虚拟机备份机制研究". 计算机工程与科学, 2009, 31(9): 131-134.