簡易檢索 / 詳目顯示
| 研究生： |
狄尚弘 Shang-Hung Ti |
|---|---|
| 論文名稱： |
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 Rethinking Virtual Memory Management to Minimize the I/O Traffic of Deep Learning Recommendation Algorithm to Open Channel SSD |
| 指導教授： |
陳增益
Tseng-Yi Chen |
| 口試委員: | |
| 學位類別： |
碩士 Master |
| 系所名稱： |
資訊電機學院 - 資訊工程學系 Department of Computer Science & Information Engineering |
| 論文出版年： | 2022 |
| 畢業學年度： | 110 |
| 語文別： | 英文 |
| 論文頁數： | 34 |
| 中文關鍵詞： | 固態硬碟 、推薦系統 、類神經網路 、虛擬記憶體 、深度學習 |
| 外文關鍵詞： | solid state drives, recommendation systems, neural networks, virtual memory, deep learning |
| 相關次數： | 點閱：12 下載：0 |
| 分享至: |
| 查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報 |
由於計算數據的快速增長，基於DRAM的主要存儲裝置無法容納來自數據密集型應用(如機器學習算法和推薦系統)的所有待處理數據。因此，主要存儲裝置和下層存儲設備之間的數據移動導致了一個重要的性能問題。當傳統的基於NAND的固態硬碟(SSD)應用於計算機架構時，性能問題無法得到解決，因為存儲驅動器無法區分來自主機系統的數據類型。然而，一種新型的存儲介質，即開放通道固態硬盤(OCSSD)，已經被提出來，提供了一條從主機端系統優化數據在存儲空間上放置的路徑。在這項研究中，我們為一個著名的數據密集型應用(即深度學習推薦系統(DLRM))在OCSSD存儲驅動器上開發了一個新的存取數據模型。我們的解決方案被稱為OC-DLRM，通過I/O單元將經常訪問的數據放在一起，可以最大限度地減少對快閃記憶體的I/O流量。根據我們的實驗結果，與傳統的虛擬內存管理方案相比，OC-DLRM明顯減少了記憶體和存儲設備之間的I/O流量。
Due to the rapid growth of computing data, DRAM-based main memory cannot accommodate all to-be-processed data from data-intensive applications (e.g., machine learning algorithms and recommendation systems). Therefore, data movement between main memory and a storage device results in a significant performance issue. When a traditional NAND-based solid-state drive (SSD) is applied to a computer architecture, the performance issue cannot be tackled because a storage drive cannot distinguish the types of data from the host system. However, a new type storage medium, namely open-channel SSD (OCSSD), has been proposed to provide a path to optimize data placement on the storage space from the host-side system. In this study, we develop a new data access model for a well-known data-intensive application (i.e., deep learning recommendation system (DLRM)) on an OCSSD storage drive. Our solution, called OC-DLRM, can minimize the I/O traffic to the flash memory storage device by considering the I/O unit of a flash memory drive to place the frequently-accessed data together. According to our experimental results, the OC-DLRM significantly decrease the amount of I/O traffic between memory and storage devices, compared with the traditional virtual memory management solution.
1. Gupta, U., et al. The architectural implications of facebook's dnn-based personalized recommendation. in 2020 IEEE International Symposium on High Performance Computer Architecture (HPCA). 2020. IEEE.
2. Zhao, Z., et al. Recommending what video to watch next: a multitask ranking system. in Proceedings of the 13th ACM Conference on Recommender Systems. 2019.
3. Zhou, G., et al. Deep interest evolution network for click-through rate prediction. in Proceedings of the AAAI conference on artificial intelligence. 2019.
4. Lui, M., et al. Understanding capacity-driven scale-out neural recommendation inference. in 2021 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS). 2021. IEEE.
5. Hui, B., et al., Personalized recommendation system based on knowledge embedding and historical behavior. Applied Intelligence, 2022. 52(1): p. 954-966.
6. Picoli, I.L., et al. Open-Channel SSD (What is it Good For). in CIDR. 2020.
7. Picoli, I.L., et al. uFLIP-OC: Understanding flash I/O patterns on open-channel solid-state drives. in Proceedings of the 8th Asia-Pacific Workshop on Systems. 2017.
8. Wang, P., et al. An efficient design and implementation of LSM-tree based key-value store on open-channel SSD. in Proceedings of the Ninth European Conference on Computer Systems. 2014.
9. Zhang, X., et al. Optimizing Performance for Open-Channel SSDs in Cloud Storage System. in 2021 IEEE International Parallel and Distributed Processing Symposium (IPDPS). 2021. IEEE.
10. González, J., et al. Application-driven flash translation layers on open-channel SSDs. in Proceedings of the 7th non Volatile Memory Workshop (NVMW). 2016.
11. Qin, H., et al., QBLKe: Host-side flash translation layer management for Open-Channel SSDs. Journal of Systems Architecture, 2021. 119: p. 102233.
12. Chen, J., et al. PATCH: Process-variation-resilient space allocation for open-channel SSD with 3D flash. in 2019 Design, Automation & Test in Europe Conference & Exhibition (DATE). 2019. IEEE.
13. Naumov, M., et al., Deep learning recommendation model for personalization and recommendation systems. arXiv preprint arXiv:1906.00091, 2019.
14. Bjørling, M., J. Gonzalez, and P. Bonnet. {LightNVM}: The Linux {Open-Channel}{SSD} Subsystem. in 15th USENIX Conference on File and Storage Technologies (FAST 17). 2017.
15. Wilkening, M., et al. RecSSD: near data processing for solid state drive based recommendation inference. in Proceedings of the 26th ACM International Conference on Architectural Support for Programming Languages and Operating Systems. 2021.
16. Wan, H., et al. FlashEmbedding: storing embedding tables in SSD for large-scale recommender systems. in Proceedings of the 12th ACM SIGOPS Asia-Pacific Workshop on Systems. 2021.
17. Kim, M. and S. Lee. Reducing tail latency of DNN-based recommender systems using in-storage processing. in Proceedings of the 11th ACM SIGOPS Asia-Pacific Workshop on Systems. 2020.
18. Soltaniyeh, M., et al. Near-Storage Processing for Solid State Drive Based Recommendation Inference with SmartSSDs®. in Proceedings of the 2022 ACM/SPEC on International Conference on Performance Engineering. 2022.
19. Eisenman, A., et al., Bandana: Using non-volatile memory for storing deep learning models. Proceedings of Machine Learning and Systems, 2019. 1: p. 40-52.
20. Paszke, A., et al., Pytorch: An imperative style, high-performance deep learning library. Advances in neural information processing systems, 2019. 32.
21. Levental, M. and E. Orlova, Comparing the costs of abstraction for DL frameworks. arXiv preprint arXiv:2012.07163, 2020.
