本研究實作並評估以 ZNS SSD 為後端儲存的可調一致性語意檔案系統，設計七種系統架構（Design A–G），涵蓋裸分區裝置(Zoned，ZNS SSD)、非分區裸裝置(Non-Zoned，Conventional SSD)、Btrfs、F2FS 及 zonefs 等後端，針對三種一致性語意（Strong、Commit、Session）、四種讀寫存取模式（File-per-process、Contiguous、Strided、Random）及兩種檔案大小（8 KB、8 MB）進行全組合效能量測。為排除快閃記憶體老化造成的效能波動，實驗統一採用 Null Block Device Driver 以 DRAM 模擬 ZNS ，並將原 TangramFS [1]的分散式叢集部署遷移至單機 TCP loopback 環境，無需伺服器叢集即可重現平行 I/O 情境。
實驗結果顯示，ZNS所引入的管理成本在大檔案（8 MB）高 I/O 量情境下對寫入頻寬影響顯著，小檔案（8 KB）情境下此差距趨於收斂；讀取頻寬在各架構間差異不顯著。在相容ZNS的檔案系統選擇上，Btrfs 在本研究環境中取得最佳寫入效能。一致性語意方面，Session 語意在小檔案讀取情境下優於 Commit，與原始文獻一致；大檔案讀取情境下 Session 語意的優勢在單機 loopback 環境中同樣成立，推測為低通訊延遲所致，與原始分散式部署結論有所差異。

This thesis explores the integration of TangramFS, a tunable-consistency parallel file system, with Zoned Namespace (ZNS) SSD storage backends, aiming to understand how the two technologies interact in practice. Seven system designs (Design A–G, shown in Figure 1) were implemented, placing TangramFS's Burst Buffer layer on raw ZNS devices, non-zoned devices, and three ZNS-compatible file systems (Btrfs, F2FS, zonefs). Write/read bandwidth and page-fault behavior were measured under three consistency semantics and four access patterns. Results show that ZNS management overhead becomes significant under large I/O but converges under small I/O; among ZNS-compatible file systems, Btrfs achieves the best write performance in this environment; and the relaxed consistency semantics of TangramFS behave consistently with the original literature under single-machine emulation.

[1] C. Wang, Parallel file system with tunable consistency, Ph.D. Dissertation, University of Illinois at Urbana-Champaign, Urbana, IL, 2022.
[2] C. Wang, K. Mohror, and M. Snir, “Formal Definitions and Performance Comparison of Consistency Models for Parallel File Systems,” IEEE Trans. Parallel Distrib. Syst., vol. 35, no. 6, pp. 1092–1106, Jun. 2024, doi: 10.1109/TPDS.2024.3391058.
[3] Y. Li and K. N. Quader, “NAND Flash Memory: Challenges and Opportunities,” Computer, vol. 46, no. 8, pp. 23–29, Aug. 2013, doi: 10.1109/MC.2013.190.
[4] N. R. Mielke, R. E. Frickey, I. Kalastirsky, et al., “Reliability of Solid-State Drives Based on NAND Flash Memory,” Proc. IEEE, vol. 105, no. 9, pp. 1725–1750, Sep. 2017, doi: 10.1109/JPROC.2017.2725738.
[5] R. Bez, E. Camerlenghi, A. Modelli, et al., “Introduction to Flash Memory,” Proc. IEEE, vol. 91, no. 4, pp. 489–502, Apr. 2003, doi: 10.1109/JPROC.2003.811702.
[6] T.-S. Chung, D.-J. Park, S. Park, et al., “A Survey of Flash Translation Layer,” J. Syst. Archit., vol. 55, no. 5, pp. 332–343, May 2009, doi: 10.1016/j.sysarc.2009.03.005.
[7] Y. Luo and M. Lin, “Flash Translation Layer: a Review and Bibliometric Analysis,” Int. J. Intell. Comput. Cybern., vol. 14, no. 3, pp. 480–508, May 2021, doi: 10.1108/IJICC-02-2021-0034.
[8] J. Kim, K. Lim, Y. Jung, et al., “Alleviating Garbage Collection Interference Through Spatial Separation in All Flash Arrays,” presented at the 2019 USENIX Annual Technical Conference (USENIX ATC 19), Renton, WA, USA, Jul. 2019, pp. 799–812.
[9] NVM Express, Inc., “NVM Command Set Specification - NVM Express.” Accessed: May 19, 2026. [Online]. Available: https://nvmexpress.org/specification/nvm-command-set-specification/
[10] M. Bjørling, “From Open-Channel SSDs to Zoned Namespaces,” in Proc. Linux Storage and Filesyst. Conf.(Valut), Boston, MA, USA, Feb. 2019, vol. 1, p. 20.
[11] M. Bjørling, H. Holmberg, A. Ramesh, et al., “ZNS: Avoiding the Block Interface Tax for Flash-based SSDs,” in USENIX Annual Technical Conference, Online, Jul. 2021, pp. 689–703.
[12] W. Qi, Z. Tan, Z. Zhang, et al., “exZNS: Extending Zoned Namespace to Support Byte-loggable Zones,” ACM Trans Arch. Code Optim, vol. 22, no. 1, p. 23:1-23:28, Mar. 2025, doi: 10.1145/3705318.
[13] J. Ramsden, When Do Zoned Namespaces SSDs Matter? : A Comparative Study of Cache Workloads, M.S. Thesis, Dept. of Electrical and Computer Engineering, University of British Columbia, Vancouver, BC, Canada, 2026. doi: 10.14288/1.0451844.
[14] N. Tehrany and A. Trivedi, “Understanding NVMe Zoned Namespace (ZNS) Flash SSD Storage Devices,” Jun. 03, 2022, arXiv: arXiv:2206.01547. doi: 10.48550/arXiv.2206.01547.
[15] Western Digital Corporation, “Zoned Storage Devices Overview | Zoned Storage.” Accessed: Apr. 19, 2026. [Online]. Available: https://zonedstorage.io/docs/introduction/zoned-storage
[16] S.-K. Wang, Enhancing File Read/Write Performance Based on Buffing I/O Mechanisms for Zoned Namespace Solid-State Drive, M.S. Thesis, Dept. of Engineering Science, National Cheng Kung University, Tainan, Taiwan, 2024.
[17] M. Oh, S. Yoo, J. Choi, et al., “ZenFS+: Nurturing Performance and Isolation to ZenFS,” IEEE Access, vol. 11, pp. 26344–26357, 2023, doi: 10.1109/ACCESS.2023.3257354.
[18] M. Lu, P. Jin, X. Wang, et al., “ZoneKV: A Space-Efficient Key-Value Store for ZNS SSDs,” in 2023 60th ACM/IEEE Design Automation Conference (DAC), San Francisco, CA, USA, Jul. 2023, pp. 1–6. doi: 10.1109/DAC56929.2023.10247926.
[19] Y. Xu, P. Jin, M. Lu, et al., “LeanKV: Efficient Garbage Collection for LSM-Based Key-Value Stores on ZNS SSDs through Lifetime-Based SSTable Clustering,” in 2023 IEEE 29th International Conference on Parallel and Distributed Systems (ICPADS), Ocean Flower Island, Hainan, China, Dec. 2023, pp. 1895–1902. doi: 10.1109/ICPADS60453.2023.00260.
[20] B. Liu, Y. Xia, X. Wei, et al., “LifetimeKV: Narrowing the Lifetime Gap of SSTs in LSMT-based KV Stores for ZNS SSDs,” in 2023 IEEE 41st International Conference on Computer Design (ICCD), Washington, DC, USA, Nov. 2023, pp. 300–307. doi: 10.1109/ICCD58817.2023.00053.
[21] N. Dryden, R. Böhringer, T. Ben-Nun, et al., “Clairvoyant Prefetching for Distributed Machine Learning I/O,” Jun. 10, 2021, arXiv: arXiv:2101.08734. doi: 10.48550/arXiv.2101.08734.
[22] C. Wang, J. Sun, M. Snir, et al., “Recorder 2.0: Efficient Parallel I/O Tracing and Analysis,” in 2020 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW), New Orleans, LA, USA (held virtually), May 2020, pp. 1–8. doi: 10.1109/IPDPSW50202.2020.00176.
[23] “IEEE Standard for Information Technology–Portable Operating System Interface (POSIX(TM)) Base Specifications, Issue 7,” IEEE Std 1003.1-2017, pp. 1–3951, Jan. 2018, doi: 10.1109/IEEESTD.2018.8277153.
[24] S. Snyder, P. Carns, K. Harms, et al., “Modular HPC I/O Characterization with Darshan,” in 2016 5th Workshop on Extreme-Scale Programming Tools (ESPT), Salt Lake City, UT, USA, Nov. 2016, pp. 9–17. doi: 10.1109/ESPT.2016.006.
[25] H. Luu, M. Winslett, W. Gropp, et al., “A Multiplatform Study of I/O Behavior on Petascale Supercomputers,” in Proceedings of the 24th International Symposium on High-Performance Parallel and Distributed Computing, in HPDC ’15. Portland, OR, USA: Association for Computing Machinery, Jun. 2015, pp. 33–44. doi: 10.1145/2749246.2749269.
[26] O. Rodeh, J. Bacik, and C. Mason, “BTRFS: The Linux B-Tree Filesystem,” ACM Trans. Storage TOS, vol. 9, no. 3, p. 9:1-9:32, Aug. 2013, doi: 10.1145/2501620.2501623.
[27] C. Lee, D. Sim, J. Hwang, et al., “F2FS: A New File System for Flash Storage,” presented at the 13th USENIX Conference on File and Storage Technologies (FAST 15), Santa Clara, CA, USA, Feb. 2015, pp. 273–286.
[28] D. L. Moal and T. Yao, “zonefs: Mapping the POSIX File System Interface to Zoned Block Device Accesses,” presented at the Proc. 2020 Linux Storage and Filesystems Conf. (VAULT’20), Santa Clara, California: USENIX, Feb. 2020.
[29] C. Hellwig, H. Holmberg, and D. Le Moal, “Staying in the Zone - Retrofitting Zoned Storage into a Scalable Enterprise File System,” in Proceedings of the 16th ACM SIGOPS Asia-Pacific Workshop on Systems, Seoul, Republic of Korea: ACM, Oct. 2025, pp. 30–37. doi: 10.1145/3725783.3764399.