Jointly Consider GC Overhead and Storage Internal Parallelism to Boost ZNS SSD Performance

以作者查詢圖書館館藏

、以作者查詢臺灣博碩士

、以作者查詢全國書目

、勘誤回報

、線上人數：10

、訪客IP：3.147.140.129

姓名

陳彥廷(Yan-Ting Chen) 查詢紙本館藏

畢業系所

資訊工程學系

論文名稱

(Jointly Consider GC Overhead and Storage Internal Parallelism to Boost ZNS SSD Performance)

相關論文

★ 重新思考虛擬記憶體管理的方式以開放通道式固態硬碟最大限度地減少深度學習推薦系統演算法的讀寫流量	★ 開啟製程相似檢查方法在組裝超級塊上以最小化額外的寫入延遲
★ LaDy: Enabling Locality-aware Deduplication Technology on Shingled Magnetic Recording Drives	★ On Minimizing Writing Overhead to Establish a Low-latency LSM-tree on Skyrmion-based Racetrack Memory
★ WABE: Rethinking B-epsilon-tree to Minimize Write-amplification on NAND Flash Memory	★ Rethinking Bϵ tree Indexing Structure over NVM with the Support of Multi-write Modes
★ Prophet’s Insight: Unleashing Deduplication System Performance in Multi-tier Storage Systems	★ Freeing the Power of High Parallelism: Accelerating the Bϵ-Tree Indexing Scheme Performance on Open-Channel SSD
★ Applying Content-Defined Chunking to OCSSD-based Deduplication Systems	★ ReLoaDing Performance: A Locality-Based Strategy for Rapid Reads in Encrypted Key-Value Systems
★ GraLoc: Preserving Graph Locality to Minimize Read and Write Amplification on NAND Flash Memory	★ On Minimizing Writing Overhead to Establish a Low-latency LSM-tree on Skyrmion-based Racetrack Memory
★ Planting a Forest in Sky: Harnessing Parallelism in Skyrmion Racetrack Memory for Efficient Random Forest Data Placement	★ LaGRange: Locality-aware Graph with Range Maintenance
★ Precision versusPerformance: Optimizing Swapping Mechanisms in Multi-Mode NVM

檔案

[Endnote RIS 格式]

[Bibtex 格式]

[相關文章]

[文章引用]

[完整記錄]

[館藏目錄]

至系統瀏覽論文 (2026-11-29以後開放)

摘要(中)

Zoned Namespace 固態硬碟（ZNS SSDs）在 Docker 和雲端計算環境中越來越被認為是關鍵組件。當前的研究主要集中在減少衝突，但未能充分利用 ZNS SSDs 所提供的並行處理能力。為了解決這一問題，我們提出了一種新的方法，通過將工作負載進行分類，並使用特定的放置策略來減少衝突。此外，我們還動態地增加並行單元的使用，以提升 ZNS SSDs 的並行性能。實驗結果顯示，與基線相比，我們的方案（稱為 VM I/O Scheduler）顯著改善了寫入延遲，平均提升 42%，讀取延遲平均提升 57%，垃圾回收（GC）開銷平均降低 61%。

摘要(英)

Zoned Namespace Solid-State Drives (ZNS SSDs) are increasingly recognized as crucial components in Docker and cloud computing environments. Current research focuses mainly on minimizing conflicts, but it does not fully take advantage of the parallelism that ZNS SSDs offer. To address this gap, we propose a new method that categorizes workloads and uses a specific placement strategy to reduce conflicts. Additionally, we dynamically increase the use of parallel units to enhance parallelism in ZNS SSDs. Our experimental results show that, compared to the baseline, our approach, called the VM I/O Scheduler, significantly improves the write latency by an average of 42%, the read latency by an average of 57% and GC overhead by an average of 61%.

關鍵字(中)

★ ZNS SSDs
★ 並行處理
★ 虛擬機
★ 性能干擾
★ 垃圾回收開銷

關鍵字(英)

★ ZNS SSDs
★ Parallelism
★ Virtual Machine
★ Performance interference
★ GC Overhead

論文目次

論文摘要 i
Abstract ii
1 Introduction 1
2 Background & Motivation 3
2.1 Zoned Namespace SSDs 3
2.2 The Application of ZNS SSDs in Docker and Cloud Environments 5
2.3 Motivation 6
3 VM I/O scheduler 8
3.1 Overview 8
3.2 Locality Table 9
3.3 Locality Detection 10
3.4 Inter-Zone Interference Detection 10
3.5 Data Placement Strategies 11
3.6 Corrective Data Reallocation 13
4 Experiments 14
4.1 Environmental Settings 14
4.2 Evaluation Result 15
4.2.1 Latency Impact 15
4.2.2 GC Overhead 15
5 Related Work 17
6 Concluding Remarks 18
Bibliography 19

參考文獻

1 Jaehong Min, Chenxingyu Zhao, Ming Liu, and Arvind
Krishnamurthy. eZNS: An Elastic Zoned Namespace for
Commodity ZNS SSDs. In Proceedings of the USENIX
Symposium on Operating Systems Design and Imple-
mentation, pages 461–477, 2023.
2 Theano Stavrinos, Daniel S Berger, Ethan Katz-Bassett, and Wyatt
Lloyd. Don’t be a blockhead: zoned namespaces make work on con-
ventional ssds obsolete. In Proceedings of the Workshop on Hot Topics
in Operating Systems, pages 144–151, 2021.
3 Matias Bjorling, Abutalib Aghayev, Hans Holmberg,
Aravind Ramesh, Damien Le Moal, Gregory R. Ganger,
and George Amvrosiadis. ZNS: Avoiding the Block
Interface Tax for Flash-based SSDs. In Proceedings of
the USENIX Annual Technical Conference, pages 689–
703, 2021.
4 Kyuhwa Han, Hyunho Gwak, Dongkun Shin, and Jooyoung Hwang.
Zns+: Advanced zoned namespace interface for supporting in-storage
zone compaction. In 15th {USENIX}Symposium on Operating Systems
Design and Implementation ({OSDI}21), pages 147–162, 2021.
5 M. Im, K. Kang, and H. Yeom, “Accelerating Rocksdb for Small-zone
ZNS SSDs by parallel I/O mechanism,” in ACM Middleware, 2022, pp.
15—-21.
6 Samsung Electronics. Multi-Stream Write SSD. Flash Memory Summit, 2016.
7 Y. Han, M. Oh, J. Lee, S. Yoo, B. S. Kim and J. Choi, "Achieving Performance Isolation in Docker Environments with ZNS SSDs," 2023 IEEE 12th Non-Volatile Memory Systems and Applications Symposium (NVMSA), Niigata, Japan, 2023, pp. 25-31.
8 R. Liu, Z. Tan, Y. Shen, L. Long and D. Liu, "Fair-ZNS: Enhancing Fairness in ZNS SSDs Through Self-Balancing I/O Scheduling," in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, vol. 43, no. 7, pp. 2012-2022, July 2024.
9 Hanyeoreum Bae, Jiseon Kim, Miryeong Kwon, and Myoungsoo Jung.
2022. What You Can’t Forget: Exploiting Parallelism for Zoned
Namespaces. In Proceedings of the 14th ACM Workshop on Hot
Topics in Storage and File Systems (HotStorage ’22). Association
for Computing Machinery, New York, NY, USA, 79–85.
10 C. Gao, L. Shi, M. Zhao, C. J. Xue, K. Wu and E. H. . -M. Sha, "Exploiting parallelism in I/O scheduling for access conflict minimization in flash-based solid state drives," 2014 30th Symposium on Mass Storage Systems and Technologies (MSST), Santa Clara, CA, USA, 2014, pp. 1-11.

指導教授

陳增益(Tseng-Yi Chen)

審核日期

2024-12-12

推文