On Minimizing Writing Overhead to Establish a Low-latency LSM-tree on Skyrmion-based Racetrack Memory

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/90026

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/90026

Title:	On Minimizing Writing Overhead to Establish a Low-latency LSM-tree on Skyrmion-based Racetrack Memory
Authors:	蕭瑋彤;HSIAO, WEI-TUNG
Contributors:	資訊工程學系
Keywords:	LSM-tree;斯格明子;複用;回收;就地;鍵值對儲存;LSM-tree;Skyrmion;Reuse;Recycling;In-place;Key-value store
Date:	2022-09-20
Issue Date:	2022-10-04 12:08:23 (UTC+8)
Publisher:	國立中央大學
Abstract:	斯格明子賽道式記憶體(Skyrmion Race Track Memory,SKRM) 具有潛力能成為未來的主流儲存裝置，然而SKRM 對現今主流的鍵值對儲存資料庫LSM-Tree(Log-Structure Merge Tree,LSM-tree) 並不友善，若直接運行LSM-tree 在SKRM 上會造成很嚴重的寫入開銷，因為斯格名子(Skyrmion,SK) 的產生和消滅是一件耗能耗時的操作。因此，為了減少SK 產生與消滅次數，在本文中提出了一個修改後對SKRM較為友善的LSM-tree，它能使用一個能在資料真正寫入到硬體前，預先計算出可能的SK 產生與消滅次數的方法，以此來決定資料的硬體寫入位置，進而降低資料寫入的成本。此修改後的LSM-tree 會復用已經存有被刪除資料的儲存區塊，並將新的資料寫入其中，以此來降低不必要的SK 產生；依據實驗結果顯示，與基線方法先比較，修改後的Lsm-tree 能降低33% 的寫入延遲並且降低68% 的寫入能耗。;The skyrmion race track memory (SKRM) has the potential to become the future mainstream memory storage device; yet, the Log-structure Merge Tree(LSM-tree) which is one of the famous databases is not friendly to run on SKRM because skyrmions creation is a high time- and energy-consuming operation. Therefore, to reduce skyrmion creation, in this work, a modified LSM-tree is proposed in using a method for pre-calculating the number of created and destroyed skyrmion before data is physically written into the SKRM hardware device. The modified LSM-tree will reuse a storage area which it already contains deleted data for writing incoming data; therefore, it can reduce the writing overhead. The final experiments show that the writing latency can be reduced by 33\% and the energy consumption can be reduced by 68\% compared with the baseline solution.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	114	View/Open

社群 sharing

Loading...