隨著雲端服務的發展以及行動裝置、物聯網裝置數量的成長，人們對於儲存空間的需求也與日俱增。而在如此需要儲存大量資料的情況下，如何去建置一個保有一定可靠度(reliability)與效能且價格低廉的大容量儲存系統(storage system)為本論文的主要研究目標。
一般而言，大多的儲存系統都建置於傳統硬碟之上，然而傳統硬碟技術在容量提升上已經遇到頻頸，無法應付不斷成長的儲存需求。因此，有了新興的磁碟技術出現，而其中最具競爭力，價格低廉的硬碟技術非 "疊瓦式磁紀錄(Shingled Magnetic Recording, SMR)硬碟" 莫屬。相較於傳統磁紀錄(Conventional Magnetic Recording, CMR)硬碟，因為SMR硬碟高儲存密度的特性，能夠達到高儲存量以及低成本的需求。同時，相較於使用CMR硬碟，使用高容量的SMR硬碟更能夠節省建置儲存系統所需要的物理空間。基於上述的理由，SMR硬碟很適合做為整個系統主要的資料儲存空間。
但由於SMR硬碟使用重疊相鄰磁軌(adjacent tracks)的方式來達到提升儲存密度(storage density)的緣故，對於隨機寫入(random write)的容忍度很低，進行資料寫入時會有寫入放大(write amplification)的問題。若單純使用SMR硬碟組成磁碟陣列以達成儲存系統的可靠性，勢必會因為頻繁更新資料(data)與同位編碼(parity)而造成系統寫入效能的瓶頸，使得系統無法提供良好的存取服務。
所以本論文提出一個基於SMR硬碟與LRC編碼的大容量儲存系統，我們利用Local Reconstruction Codes(LRC)作為儲存系統的容錯(fault tolerance)機制，因為LRC資料的重建(Reconstruction)成本比傳統Reed Solomon codes(RS codes)的方式低許多，容錯能力也相對比較好。同時，我們採用SMR與CMR硬碟混用的架構，使系統在降低成本的情況下保有一定的效能。而為了要更進一步提升儲存效能，本論文設計了一套資源管理機制，以管理資料與同位編碼的儲存與更新，來減少因採用SMR硬碟所造成的效能下降問題。
實驗數據顯示本論文所提出的磁碟架構相較於單純的SMR磁碟陣列能大大降低資料寫入的時間，而透過提出的管理機制則讓系統的效能更加提升。

The demand for storage space grows as cloud services, vast amounts of mobile and IoT devices emerge. To provide high capacity with reasonable reliability and performance while keeping the low cost is the major purpose of this research.
In general, conventional magnetic recording drives are used to construct storage systems; however, the conventional disk drives cannot handle the non-stopping demand for storage space due to the physical limitation of itself. Therefore, some emerging magnetic recording techniques are proposed in recent years and Shingled Magnetic Recording technique is the most promising of them due to its low cost. SMR disk drives can provide higher capacity with lower cost compared to that of conventional disk drives. Hence, use SMR drives as the major storage components in the large storage system is a good solution. However, SMR drives overlapped adjacent tracks to increase storage density, which leads to write amplification problem. Utilize SMR drives as storage components in RAID to build to a fault-tolerant storage system can lower the system performance greatly, since updating parity frequently can even degrade the write performance.
Therefore, in this thesis, a high capacity storage system based on SMR drives and Local Construction Codes is proposed. First, in the proposed system,  LRC is applied as the fault-tolerant mechanism, because of that the system reconstruction overhead of LRC is lower than that of Reed Solomon Codes when a disk failure happened. Then, SMR drives  and CMR drives are utilized to store data and parity, respectively for enhancing the system performance and decrease the cost. Finally, a resource management mechanism is proposed to handle the read, write, and update request of data and of parity to enhance the overall access performance.
The simulation results show that the proposed system can greatly improve the access performance compared to pure SMR RAID system. In addition, the proposed resource management mechanism can further reduce the write amplification effect of SMR drives and enhance the system efficiency.

目錄
目錄	IV
圖目錄	VII
表目錄	IX
第一章	緒論	1
1.1	前言	1
1.2	動機與目的	2
1.3	論文章節架構	4
第二章	背景知識與相關文獻	5
2.1	Shingled Magnetic Recording (SMR)	5
2.2	SMR的種類	9
2.2.1	Drive-Managed (DM)	9
2.2.2	Host-Managed (HM)	10
2.2.3	Host-Aware (HA)	11
2.3	RAID與LRC	12
2.3.1	Redundant Array of Independent Disks (RAID)	12
2.3.2	Local Reconstruction Codes (LRC)	14
2.4	SMR硬碟的容錯機制	17
2.4.1	RAID 4S	18
2.4.2	SWD-based RAID system	19
2.4.3	Hybrid SMR RAID	21
第三章	基於LRC之高容錯儲存系統	23
3.1	系統架構	24
3.2	磁碟陣列存取方式	27
3.3	資料更新方式	29
3.3.1	附加更新(Appended Update)	29
3.3.2	反向更新(Reversed Update)	30
3.4	系統流程圖	35
第四章	實驗結果與分析	41
4.1	模擬環境	41
4.1.1	真實與模擬的CMR硬碟比較	41
4.1.2	真實與模擬的RAID 6比較	43
4.2	實驗數據與分析	44
4.2.1	LRC組成硬碟種類之效能比較	45
4.2.2	調整更新比例對於系統的影響	51
4.2.3	調整更新資料集中度對於系統的影響	53
第五章	貢獻與未來展望	56
5.1	主要貢獻	56
5.2	未來展望	56
參考文獻	57

 
圖目錄
圖 1.1	IDC對於日後資料領域使用量的預測	1
圖 1.2	各廠硬碟修復效率比較[6]	3
圖 2.1	CMR與SMR之示意圖比較	5
圖 2.2	CMR與SMR隨機寫入示意圖	8
圖 2.3	RMW示意圖	9
圖 2.4	常見RAID示意圖	14
圖 2.5	RS (6, 3)示意圖	15
圖 2.6	LRC (6, 2, 2)示意圖	16
圖 2.7	LRC和RS的儲存開銷成本與重建成本比較圖[10]	17
圖 2.8	RAID 4S系統架構	19
圖 2.9	SWD-based RAID系統架構	20
圖 2.10	Hybrid SMR RAID 系統架構示意圖	22
圖 3.1	系統所在之層級	24
圖 3.2	系統架構圖	25
圖 3.3	磁碟陣列儲存架構	28
圖 3.4	附加更新示意圖	30
圖 3.5	反向更新	32
圖 3.6	小檔案更新示意圖	35
圖 3.7	大檔案更新示意圖	35
圖 3.8	系統流程圖	39
圖 3.9	SMR寫入流程圖	40
圖 4.1	真實與模擬的CMR硬碟反應時間的累積分布圖	42
圖 4.2	真實與模擬的RAID 6反應時間的累積分布圖	44
圖 4.3	使用全CMR之LRC儲存系統延遲時間的累積分布圖	47
圖 4.4	使用全SMR之LRC儲存系統延遲時間的累積分布圖	48
圖 4.5	LRC儲存系統反應時間的累積分布圖	50
圖 4.6	調整更新比例對於系統的影響	53
圖 4.7	調整更新資料集中度對於系統的影響	55

 
表目錄
表 3.1	stripe_head結構	28
表 3.2	au_mt結構	30
表 3.3	ru_mt結構	32
表 3.4	temp_list結構	33
表 3.5	熱度公式參數	33
表 4.1	硬碟基本資訊	41
表 4.2	量測參數	43
表 4.3	使用全CMR的LRC之實驗參數	45
表 4.4	使用全SMR的LRC儲存系統之實驗參數	47
表 4.5	LRC儲存系統之實驗參數	49
表 4.6	所有系統平均延遲之比較	51
表 4.7	調整更新比例之實驗參數	51
表 4.8	調整更新資料集中度之實驗參數	54

[1]	Alex Woodie, "Global DataSphere to Hit 175 Zettabytes by 2025, IDC Says", 2015, [online] Available: https://www.datanami.com/2018/11/27/global-datasphere-to-hit-175-zettabytes-by-2025-idc-says/
[2]	Thompson, D. A., & Best, J. S. (2000). “The future of magnetic data storage technology.” IBM Journal of Research and Development, 44(3), 311-322.
[3]	Y. Shiroishi, K. Fukuda, I. Tagawa, S. Takenoiri, H. Tanaka, and N. Yoshikawa. “Future options for HDD storage.”,  IEEE Transactions on Magnetics, vol. 45, no. 10, pp. 3816-3822, Oct. 2009.
[4]	Anton Shilov, “Western Digital: Over Half on Data Center HDDs Will Use SMR by 2023” , 2019, https://www.anandtech.com/show/14099/western-digital-over-half-of-dc-hdds-will-use-smr-by-2023
[5]	ZFS on Linux, .[online] Available: https://zfsonlinux.org/
[6]	Joel Hruska, “Red Alert: WD Sued for Selling ‘Inferior’ SMR Hard Drives to NAS Customers”, 2020, [online] Available: https://www.extremetech.com/computing/311182-red-alert-wd-sued-for-selling-inferior-smr-hard-drives-to-nas-customers
[7]	Shingled Magnetic Recording, [online] Available: https://zonedstorage.io/introduction/smr/
[8]	Margaret Rouse, “erasure coding”, 2014, [online] https://searchstorage.techtarget.com/definition/erasure-coding
[9]	RAID, [online] Available: https://en.wikipedia.org/wiki/RAID
[10]	Reed–Solomon error correction, [online] Available: https://en.wikipedia.org/wiki/Reed%E2%80%93 Solomon_error_correction
[11]	A. Fikes, “Storage Architecture and Challenges,” Google Faculty Summit, 2010.
[12]	D. Ford et al., “Availability in Globally Distributed Storage Systems,”, USENIX OSDI, 2010.
[13]	D. Borthakur et al., “HDFS RAID,” Hadoop User Group Meeting, Nov. 2010.
[14]	Quoc Minh Le, Ahmed Amer, and JoAnne Holliday. 2015. SMR disks for mass storage systems. In 2015 IEEE 23rd International Symposium on Modeling, Analysis, and Simulation of Computer and Telecommunication Systems. 228–231.
[15]	I. Adams, B. Madden, J. Frank, M. W. Storer, E. L. Miller and G. Harano, "Usage behavior of a large-scale scientific archive", Proceedings of the 2012 International Conference for High Performance Computing Networking Storage and Analysis (SC12), Nov. 2012.
[16]	W. Liu et al., “Understanding the SWD-based RAID system,” in CCBD, 2014.
[17]	Lu, Z. W. and G. Zhou, “Design and implementation of hybrid shingled recording RAID system,” 2016 IEEE 14th Intl. Conf. on Dependable, Autonomic and Secure Computing, 14th Intl. Conf. on Pervasive Intelligence and Computing, 2nd Intl. Conf. on Big Data Intelligence and Computing and Cyber Science and Technology Congress (DASC/PiCom/DataCom/CyberSciTech), 937–942, Auckland, 2016.
[18]	I. T. T. Committee, Information Technology-Zoned Block Commands (ZBC), 2016, [online] Available: http://www.t10.org/.
[19]	I. T. T. Committee, Information Technology Zoned Device ATA Command Set (ZAC), 2016, [online] Available: http://www.t13.org/.
[20]	T. Yao et al., "SEALDB: An efficient LSM-tree based KV store on SMR drives with sets and dynamic bands", IEEE Trans. Parallel Distrib. Syst., vol. 30, no. 11, pp. 2595-2607, Nov. 2019.