基于闪存的数据缓存关键技术研究

项目名称： 基于闪存的数据缓存关键技术研究

项目编号： No.61303040

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 汤显

作者单位： 燕山大学

项目金额： 25万元

中文摘要： 虽然闪存以其小巧、轻便、无机械延迟等特点被认为有望替代磁盘作为未来的二级存储设备，但是基于闪存的数据管理技术远未发展成熟。在未来相当长一段时间内，根据不同应用的特点和要求，实际中将出现各种基于闪存的存储系统。在这些系统中，缓冲区是其核心组件，缓冲区置换算法的高效与否直接决定了系统的效率。虽然研究者已经提出了各种基于闪存的缓冲区置换策略，但其基本假设是闪存的随机读代价远小于随机写代价。这一假设和实际不符，实际中不但单一闪存存在读写代价不对称的问题，而且不同型号闪存读写不对称的程度之间存在巨大差异性。本课题基于这一背景，重点研究基于闪存的缓冲区管理中涉及的关键技术问题，包括单介质、多介质、多级缓冲区管理机制以及混合式系统中的数据放置策略。本项目的研究有助于系统阐明在各种基于闪存的存储系统中的自适应置换策略、相关算法及模型，具有重要的理论意义和应用价值。

中文关键词： 闪存；数据库；缓冲区；存储模型；置换策略

英文摘要： Flash disks are considered as important alternatives to conventional magnetic disks because of their distinguished feature, such as being small, lightweight and no mechanical delay etc. However, flashed based data management techniques are still far from ideality. In the long future, there will be many flashed based storage system that are constructed in terms of the unique requirements of various applications. In all these systems, buffer is one of the most fundamental components, and any substantial progress in buffer replacement algorithms will affect the entire modern computational stack. Although researchers have proposed many flash-aware buffer replacement algorithms, their assumption is that the cost of random read operation is neglectable compared with that of random write operation, which does not consistent with the actuality. In fact, besides the fact that the cost of random read operation is less than that of random write operation, a more important thing is that there exists huge discrepancy to the ratio of write cost to read cost on different flash disks. In this project, we focus on making improvements on the key technologies related to buffer management for flash-based system, include buffer management strategy against single-storage media, multi-types of storage media, and the data placement in

英文关键词： Flash；Database；Buffer；Storage Model；Replacement Strategy