基於持續性記憶體的硬體加速預寫式日誌

Hardware-Assisted Write-Ahead Logging for Persistent Memory

董乃嘉(Nai-Jia Dong)

指導教授 ： 楊佳玲

共同指導教授 ： 鄭湘筠(Hsiang-Yun Cheng)

國立臺灣大學/電機資訊學院/資料科學學位學程/碩士(2020年)

https://doi.org/10.6342/NTU202004369

摘要

持續性記憶體前景看好，因為它可以提供與 DRAM 相當的性能，並同時具有儲存裝置的持久性。
隨著 Intel Optane DIMM 發布，持續性記憶體的採用已成為現實的解決方案。
持續性記憶體的挑戰是如何有效地支援原子持久性。
在這篇論文中，我們提出了第一個基於 Intel Optane DIMM 架構的硬體加速日誌。
我們提出的複合式日誌機制是基於 Intel Optane DIMM 的兩個新功能，記憶體控制器中的持續性 WPQ 和 DIMM 上的控制器而設計的。
我們利用 WPQ 作為放置重做日誌的區域，並擴展 DIMM 控制器以支援持續性記憶體內複製，因此撤消日誌可以在持續性記憶體內部進行，而不會引起額外的記憶體匯流排流量。
結果表明，提出的機制是實用且有效的（與 Intel Optane DIMM 相比，硬體開銷微不足道），與理想情況下（即不保證一致性）相比，事務吞吐量僅略低於 3.8%，並且與 LAD 相比，吞吐量高於 41.0%。

關鍵字

持續性記憶體 ； 非揮發性記憶體 ； 崩潰一致性 ； 原子持久性 ； 硬體加速日誌 ； 預寫式日誌 ； 重做日誌 ； 撤銷日誌 ； 複合式日誌

並列摘要

Persistent memory (PM) is promising as it offers comparable performance to DRAM with the durable property of storage devices.
The adoption of PM has has become a real-world solution with the release Intel Optane DIMM.
The key challenge for persistent memory is to support atomic durability efficiently.
In this thesis, we propose the first hardware-based logging mechanism based on Intel Optane DIMM architecture.
The proposed hybrid in-PM Hybrid logging mechanism is designed based on two new features of Intel Optane DIMM, persistent write pending queues in the memory controller, and the DIMM controller.
We leverage the persistent write pending queue to serve as a redo log area and augment DIMM controller to support in-memory copy so undo logging can proceed inside PMs without introducing extra bus traffics.
The results show the proposed logging mechanism is practical (insignificant hardware overheads compared Intel Optane DIMM) yet effective, within 3.8% transaction throughput of the ideal case (without consistency guarantee) and 41.0% higher throughput than LAD.

並列關鍵字

persistent memory ； non­-volatile memory ； crash consistency ； atomic durability ； hardware-assisted logging ； write-ahead logging ； redo logging ； undo logging ； in-pm hybrid logging

參考文獻

[1] N. Binkert, B. Beckmann, G. Black, S. K. Reinhardt, A. Saidi, A. Basu, J. Hestness, D. R. Hower, T. Krishna, S. Sardashti, R. Sen, K. Sewell, M. Shoaib, N. Vaish, M. D. Hill, and D. A. Wood. The gem5 simulator. SIGARCH Comput. Archit. News, 39(2):1–7, Aug. 2011.

[2] M. Cai, C. C. Coats, and J. Huang. Hoop: Efficient hardware-assisted out-of-place update for non-volatile memory. In 2020 ACM/IEEE 47th Annual International Symposium on Computer Architecture (ISCA), pages 584–596, 2020.

[3] N. Chatterjee, N. Muralimanohar, R. Balasubramonian, A. Davis, and N. P. Jouppi. Staged reads: Mitigating the impact of dram writes on dram reads. In IEEE International Symposium on High-Performance Comp Architecture, pages 1–12, 2012.

[4] Y. Choi, I. Song, M. Park, H. Chung, S. Chang, B. Cho, J. Kim, Y. Oh, D. Kwon, J. Sunwoo, J. Shin, Y. Rho, C. Lee, M. G. Kang, J. Lee, Y. Kwon, S. Kim, J. Kim, Y. Lee, Q. Wang, S. Cha, S. Ahn, H. Horii, J. Lee, K. Kim, H. Joo, K. Lee, Y. Lee, J. Yoo, and G. Jeong. A 20nm 1.8v 8gb pram with 40mb/s program bandwidth. In 2012 IEEE International Solid-State Circuits Conference, pages 46–48, 2012.

[5] K. Doshi, E. Giles, and P. Varman. Atomic persistence for scm with a non-intrusive backend controller. In 2016 IEEE International Symposium on High Performance Computer Architecture (HPCA), pages 77–89, 2016.