DB-Tree:一种高性能的闪存数据库索引结构  

DB-TREE:AN EFFICIENT INDEX STRUCTURE FOR FLASH-BASED DATABASE

作　　者：房俊华[1] 王翰虎[1,2] 陈梅[1] 马丹[1] 

机构地区：[1]贵州大学计算机科学与信息学院,贵州贵阳550025 [2]贵州星辰科技开发有限公司,贵州贵阳550025

出　　处：《计算机应用与软件》2013年第11期243-246,共4页Computer Applications and Software

基　　金：贵阳市2010年工业科技攻关项目([2010]筑科工合同字第28号)

摘　　要：闪存性能的优势使得闪存数据库系统成为目前研究的一个热点,索引是提高闪存数据库效率的一个重要手段。基于B+树索引结构,提出一种适用于闪存数据库的索引方法:DB-Tree。该方法将更新操作以一棵"伪B+树"的结构形式存储来避免检索时扫描整个更新日志区;以分支合并的方式使更新操作有针对性地聚集于闪存页;引入更新缓冲区大小及合并频率的自适应机制使闪存数据库适用于不同的读写负载。通过与经典的日志更新IPL B+TREE及无日志的μ-Tree索引方法的实验比较,证明所提出的DB-Tree在有效降低更新代价的同时大幅度提高了索引的查询性能。The advantage of flash memory performance brings the flash-baSed database system to the focus of current research, and the index is an important means to improve the efficiency of flash-based database. In this paper, we propose a suitable index structure for flash-based database, namely DB-Tree, based on B ＋- tree index structure. In the method, the updating operations are stored in a structure form of ＂pseudo B ＋- tree＂ to avoid scanning the entire update log-area while retrieving; the update operations are pertinently aggregated to the page of flash memory in the way of merging the corresponding branches ; the self-adaptive mechanism for update-buffer size and merger frequency is introduced to ensure the flash-based database can be applied to different read/write load. Through the experimental comparison with the classic log update IPL B ＋ Tree and the μ-Tree index method without log, it is proved that the DB-Tree index structure we presented significantly reduces the update cost and greatly improves the query performance of the index simultaneously.

关 键 词：闪存 数据库 索引结构 检索 更新合并 

分 类 号：TP311.12[自动化与计算机技术—计算机软件与理论]