硬碟的單位面積儲存容量已經快達到物理上的限制，而疊瓦式磁紀錄可以使用現有硬碟技術、並快速解決該問題的新科技。在疊瓦式磁紀錄硬碟中、因為循序寫入限制的緣故，通常會配置一個永久快取(或稱作日誌型快取)以避免大量的讀取-修改-寫回的操作造成的寫入放大。在硬碟上使用循序寫入的日誌型快取有其限制，因此我們需要一個針對循序寫入特性所設計的日誌型快取，以增進疊瓦式磁紀錄硬碟的使用效能。
在這篇論文中，我們提出一個針對疊瓦式磁紀錄硬碟特性所設計的日誌型快取，在空間管理上、我們劃分為兩部分：靜態配置的固定快取以及動態配置的自由快取，透過自由快取的分配、我們能針對所需的位址分配更多空間，並延長熱資料留在快取內的時間，以達到減少垃圾回收的目的；而在垃圾回收設計的機制上、我們能在每次垃圾回收時都在最短時間內清出最多空間。
我們實作我們的日誌型快取設計在Kernel module所模擬的疊瓦式磁紀錄硬碟上，實驗結果顯示在垃圾回收的相關測量條件下、我們的設計都能有效改善垃圾回收的效率與減少額外的負荷，而在實際上的效能表現、我們的方法在不同的負載應用下、平均能降低19%的垃圾回收延遲與17%的平均寫入延遲。

The areal density of hard disks is reaching the limit. Shingled magnetic recording (SMR) is one of the promising solutions that further extends the areal density but also inherits existing hard disk technology. The design nature of SMR introduces write amplification due to the read-modify-write (RMW) operation when writing on overlapping tracks. In order to lower the impact of write amplification, persistent cache is often used on drive-managed to avoid frequent RMW operations. The cache with sequential-write property on SMR disks needs a new solution to improve SMR disks performance.
In this dissertation, we propose a log-based cache design for sequential-write-constrained properties of SMR disks. The cache is divided into two parts, static cache for static allocation and free cache for dynamic allocation. The free cache is dynamically allocated for the data to be written when the static cache is full, thus extend the period of time hot data stay in the cache and reduce the time taken by garbage collection (GC). The log-based cache design also uses cost-benefit strategy during GC to perform most efficient reclaim.
We implement log-based cache design in a kernel-based SMR disk simulator. The experiments demonstrate an improvement on the GC performance and a reduction on the overhead. The log-based cache design improves GC latency by 19% and average write latency by 17% on average under different workload.

1. 簡介 1
2. 背景介紹 3
2.1 疊瓦式磁紀錄硬碟 3
2.2 動機 5
3. 相關文獻 8
4. 日誌型快取設計 10
4.1 概述 10
4.2 空間分配策略 11
4.3 垃圾回收策略 13
4.3.1 硬碟延遲計算 14
5. 評估 16
5.1 實作 16
5.2 Trace資訊 17
5.3 比較方法 17
5.4 實驗結果 18
5.3.1 垃圾回收的帶數量 19
5.3.2 寫入放大 20
5.3.3 垃圾回收延遲 21
5.3.2 平均延遲 21
6. 結論與未來研究方向 23
7. 參考資料 24

[1] 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, Oct. 2009.
[2] G. Gibson and G. Ganger. Principles of operation for shingled disk devices. In Carnegie Mellon University Parallel Data Lab, Tech. Rep. CMU-PDL-11-107, April 2011.
[3] Garth Gibson and Milo Polte. Directions for shingled-write and two-dimensional magnetic recording system architectures: Synergies with solid-state disks. Technical report, CMU Parallel Data Laboratory, May 2009. Technical Report CMU-PDL-09-104.
[4] INCITS T10 Technical Committee. Information technology - Zoned Block Commands (ZBC). Draft Standard T10/BSR INCITS 536, American National Standards Institute, Inc., Sept. 2014. Available from http://www.t10.org/drafts.htm.
[5] Y. Cassuto, M. A. A. Sanvido, C. Guyot, D. R. Hall, and Z. Z. Bandic. Indirection Systems for Shingled-recording Disk Drives. In Proceedings of the 2010 IEEE 26th Symposium on Mass Storage Systems and Technologies (MSST), MSST ’10, pages 1–14, Washington, DC, USA, 2010. IEEE Computer Society.
[6] D. Hall, J. H. Marcos, and J. D. Coker. Data Handling Algorithms For Autonomous Shingled Magnetic Recording HDDs. IEEE Transactions on Magnetics, 48(5):1777–1781, 2012.
[7] Weiping He and David H.C. Du. SMaRT: An Approach to Shingled Magnetic Recording Translation. 15th USENIX Conference on File and Storage Technologies (FAST 17), 121—134 ,2017
[8] Abutalib Aghayev and Theodore T’so and Garth Gibson and Peter DesnoyersEvolving. Ext4 for Shingled Disks. 15th USENIX Conference on File and Storage Technologies (FAST 17), 105—120, 2017
[9] W. He and D. H. Du. Novel address mappings for shingled write disks. In 6th USENIX Workshop on Hot Topics in Storage and File Systems (HotStorage 14). USENIX Association.
[10] R. Pitchumani, J. Hughes, and E. L. Miller. Smrdb: key-value data store for shingled magnetic recording disks. In Proceedings of the 8th ACM International Systems and Storage Conference, page 18. ACM, 2015.
[11] Seagate Archive HDD. http://www.seagate.com/products/enterprise-servers-storage/nearline-storage/archive-hdd/.
[12] WD UltraStar Ha10. http://www.hgst.com/products/hard-drives/ultrastar-archive-ha10.
[13] Abutalib Aghayev and Peter Desnoyers. Skylight—a window on shingled disk operation. In 13th USENIX Conference on File and Storage Technologies, pages 135–149, 2015.
[14] F. Wu, M.-C. Yang, Z. Fan, B. Zhang, X. Ge, and D. H. Du. Evaluating host aware smr drives. In 8th USENIX Workshop on Hot Topics in Storage and File Systems (HotStorage 16). USENIX Association, 2016.
[15] B. S. Parsons and V. S. Pai, "A mathematical hard disk timing model for full system simulation," 2013 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS), Austin, TX, 2013, pp. 143-153.
[16] I. TAGAWA and M. WILLIAMS. High density data-storage using shingle-write. In Proceedings of the IEEE International Magnetics Conference, 2009.
[17] S. Greaves, Y. Kanai, and H. Muraoka. Shingled recording for 23tbit/in2. In IEEE Transactions on Magnetics, volume 45, pages 3823–3829, September 2009