傳統硬碟之儲存區域密度成長因物理限制而達到臨界點，而疊瓦式硬碟是可突破此物理限制方法之一，藉由交疊磁軌僅留下足以讀取之寬度來提升區域密度。因該機制致使在疊瓦式硬碟中寫入動作受到限制，因此需對於資料映射做進一步之設計。與疊瓦式轉換層相似之設計為快閃記憶體轉換層，同樣皆是以為克服儲存裝置之寫入限制為主旨，其中DFTL是平均效率最佳且穩定之設計，因此透過參考其架構，針對疊瓦式硬碟加以優化改良。並且考慮資料特性之行為，針對各類型採取對應之管理行為。
本篇論文提出一為解決疊瓦式磁紀錄之隨機寫入問題之Drive-management類型之疊瓦式轉換層設計。此法會將映射資訊儲存至LRU Interval Tree與Local Translation Page兩部分，並透過存放於記憶體之LRU Interval Tree來加速讀取速度。對於資料特性部分，該設計會將所有資料分為Hot/Cold兩種類型，並且利用兩者之更新頻率不同給予不同分佈之Data Band，並且依照存取行為與現今硬碟狀態設計四種垃圾回收機制，藉此提升寫入速度，並同時增加系統之Band分配靈活性。透過實驗結果可得知，此方法之存取速度甚優於傳統疊瓦式硬碟所採用之RMW，儘管在極端情況IOPS仍保持在223，此外對於系統之穩定性也有所提升。

The capacity growth of conventional magnetic hard disk (HDDs) slackens due to the physical limitation of areal density. Shingled Magnetic Recording (SMR) is one of the technologies for overcoming this predicament. However, the overlapped tracks layout in Shingles Magnetic Recording incurs write amplification due to the read-modify-write operation. Drive-managed SMR drives is proposed for this problem by maintaining translation page. Two major issues of Drive-managed SMR are performance of writing operation and overhead of garbage collection.
In this paper, we present an adaptive address mapping Shingled Translation Layer (STL) based on data feature. Translation information is stored in two-layer architecture, LRU interval tree and local translation layer. LRU interval tree in memory contains least recently used translation information for improving read operation. Data feature (Hot/Cold) represents the access and update frequency of data. Therefore, we utilize data feature as data allocating norm for separated management mechanics and garbage collection policies. The design of garbage collection guarantees the system flexibility and reduce overhead. We implement DSTL in Disksim disk simulator. The experiments results demonstrate the improvement of performance in DSTL compared to RMW and stability compared to HDDs.

Chapter 1. 介紹 1
Chapter 2. 背景與研究動機 3
Chapter 3. 相關文獻 9
Chapter 4. 研究方法 10
4.1. 概述 10
4.2. DSTL架構 11
4.2.1. LRU Interval Tree 11
4.2.2. Band Situation 11
4.2.3. Global Translation Directory (GTD) 12
4.2.4. Data Band 12
4.2.5. Interval Tree 14
4.3. 讀寫操作 15
4.3.1. 動態配置Band Type 15
4.3.2. Translation Layer運作方式 15
4.4. 垃圾回收策略 18
Chapter 5. 實驗結果 22
5.1. 實驗環境設定 22
5.2. 數據結果 23
5.2.1. LRU Elimination 23
5.2.2. 垃圾回收 24
Chapter 6. 結論與未來研究方向 29
Reference 30

[1] Thompson, David A., and John S. Best. "The future of magnetic data storage technology." IBM Journal of Research and Development 44.3 (2000): 311.
[2] Shiroishi, Y., et al. "Future options for HDD storage." IEEE Transactions on Magnetics 45.10 (2009): 3816-3822.
[3] Tagawa, Ikuya. "High density data-storage using shingled-write." INTERMAG 2009, Sacramento, May (2009).
[4] Greaves, Simon, Yasushi Kanai, and Hiroaki Muraoka. "Shingled Recording for 2–3 Tbit/in2." IEEE Transactions on Magnetics 45.10 (2009): 3823-3829.
[5] Feldman, Tim, and Garth Gibson. "Shingled magnetic recording: Areal density increase requires new data management." USENIX; login: Magazine 38.3 (2013).
[6] Aghayev, Abutalib, Mansour Shafaei, and Peter Desnoyers. "Skylight—a window on shingled disk operation." ACM Transactions on Storage (TOS) 11.4 (2015): 16.
[7] Chinner, D. "SMR Layout Optimization for XFS." (2015).
[8] Jin, Chao, et al. "HiSMRfs: A high performance file system for shingled storage array." Mass Storage Systems and Technologies (MSST), 2014 30th Symposium on. IEEE, 2014.
[9] Wu, Fenggang, et al. "Evaluating Host Aware SMR Drives." HotStorage. 2016.
[10] Gibson, Garth, and Greg Ganger. "Principles of operation for shingled disk devices." USENIX HotStorage (2011).
[11] Gibson, Garth, and Milo Polte. "Directions for shingled-write and two-dimensional magnetic recording system architectures: Synergies with solid-state disks." Parallel Data Lab, Carnegie Mellon Univ., Pittsburgh, PA, Tech. Rep. CMU-PDL-09-014(2009).
[12] Gupta, Aayush, Youngjae Kim, and Bhuvan Urgaonkar. DFTL: a flash translation layer employing demand-based selective caching of page-level address mappings. Vol. 44. No. 3. ACM, 2009.
[13] Ban, Amir. "Flash-memory translation layer for NAND flash (NFTL)." M-systems (1998).
[14] Kim, Jesung, et al. "A space-efficient flash translation layer for CompactFlash systems." IEEE Transactions on Consumer Electronics 48.2 (2002): 366-375.
[15] Lee, Sang-Won, et al. "A log buffer-based flash translation layer using fully-associative sector translation." ACM Transactions on Embedded Computing Systems (TECS) 6.3 (2007): 18.
[16] Bucy, J. S. "DiskSim 4.0 (CMU-PDL-08-101)." Parallel Data Laboratory, Tech. Rep (2008).
[17] MSR Cambridge Block I /O Traces. http://iotta.snia.org/traces/388
[18] Hall, David, John H. Marcos, and Jonathan D. Coker. "Data Handling Algorithms for Autonomous Shingled Magnetic Recording HDDs." IEEE Transactions on Magnetics 48.5 (2012): 1777-1781.
[19] Cassuto, Yuval, et al. "Indirection systems for shingled-recording disk drives." Mass Storage Systems and Technologies (MSST), 2010 IEEE 26th Symposium on. IEEE, 2010.
[20] Jin, Chao, et al. "HiSMRfs: A high performance file system for shingled storage array." Mass Storage Systems and Technologies (MSST), 2014 30th Symposium on. IEEE, 2014.
[21] Luo, Dan, et al. "Design and implementation of a hybrid shingled write disk system." IEEE Transactions on Parallel and Distributed Systems 27.4 (2016): 1017-1029.
[22] Jones, Stephanie N., et al. "Classifying data to reduce long-term data movement in shingled write disks." ACM Transactions on Storage (TOS) 12.1 (2016): 2.