组合盘GRACE散列连接算法

发布时间：2018-05-20 23:06

本文选题：固态盘缓存 + 随机I/O　；参考：《小型微型计算机系统》2014年03期

【摘要】：在传统数据库系统中,I/O开销一直是连接算法性能瓶颈.固态盘具有随机读写性能高和低延迟的特性,使得利用固态盘改善连接算法I/O性能成为可能.提出基于硬盘和固态盘混合存储系统下的组合盘GRACE散列连接算法(CGHJ).CGHJ对传统GRACE散列连接算法分区阶段作如下改进:将固态盘作为内存和硬盘之间的缓存,位于内存的各个散列桶先写到位于固态盘的临时文件缓存;临时文件缓存满后,利用迁移缓存将固态盘中的内容转移到位于硬盘的各个分区文件.新的分区方式增大了分区过程中随机写到每个分区文件的块,减少硬盘分区过程中产生的随机I/O.实验结果显示在缓存容量较小或表的数据规模较大情况下,CGHJ能取得很好消除硬盘随机I/O的效果.
[Abstract]:In the traditional database system, the I/O overhead is always the bottleneck of the performance of the connection algorithm. The solid-state disk has the characteristics of high random reading and writing performance and low delay. It makes it possible to improve the performance of the connection algorithm I/O using the solid disk. A combination disk GRACE hash connection algorithm (CGHJ).CGHJ based on the hard disk and solid disk hybrid storage system is proposed to the traditional GRA. The following improvements are made in the partitioning phase of the CE hash connection algorithm: the cache of the solid disk as between the memory and the hard disk, each hash in memory first to the temporary file cache in the solid disk; after the temporary file is cached, the content of the solid disk is transferred to the various partition files located on the hard disk by the migration cache. The result of random I/O. experiment produced during partition of hard disk shows that CGHJ can effectively eliminate the effect of random I/O of hard disk when the cache capacity is small or the data scale of the table is large.
【作者单位】：浙江工业大学计算机科学与技术学院;浙江省可视媒体智能处理技术研究重点实验室;
【基金】：国家自然科学基金项目(61070042)资助浙江省自然科学基金项目(Y1090096,Y13F020114)资助
【分类号】：TP333;TP311.13

【参考文献】