一种基于裸闪存的Key-Value数据库优化方法

发布时间：2018-01-07 22:19

本文关键词：一种基于裸闪存的Key-Value数据库优化方法　出处：《计算机研究与发展》2017年06期 　论文类型：期刊论文

【摘要】：近年来,非关系型的key-value数据库得到越来越广泛的应用.然而,目前主流的key-value数据库或者是基于磁盘设计的,或者是传统的基于文件系统和闪存转换层FTL来构建的,难以发挥闪存存储设备的特性,限制了I/O的并发性能,且垃圾回收过程复杂.设计并实现了一种基于裸闪存的keyvalue数据管理架构Flashkv,通过用户态下的管理单元进行空间管理和垃圾回收,充分利用了闪存设备内部的并发特性,并简化了垃圾回收过程,去除了传统文件系统和FTL中的冗余功能,缩短了I/O路径.提出了基于闪存特点的I/O调度技术,优化了闪存的读写延迟,提高了吞吐率;提出了用户态缓存管理技术,降低了数据写入量和频繁系统调用所带来的开销.测试结果表明,Flashkv性能是levelDB的1.9~2.2倍,写入量减少60%~65%.
[Abstract]:In recent years, non-relational key-value database has been more and more widely used. However, the mainstream key-value database is based on disk design. Or the traditional FTL based on file system and flash memory conversion layer, it is difficult to play the characteristics of flash storage device, which limits the concurrent performance of I / O. And the garbage collection process is complex. This paper designs and implements a keyvalue data management architecture based on bare flash memory, Flashkv. it manages the space and garbage collection through the management unit in the user state. It makes full use of the concurrency characteristics of flash memory devices, simplifies the garbage collection process, and removes the redundant functions in traditional file systems and FTL. The I / O path is shortened. I / O scheduling technology based on flash memory is proposed to optimize the read / write delay of flash memory and improve the throughput. The user state cache management technology is proposed to reduce the overhead caused by data writing and frequent system calls. The test results show that the performance of Flashkv is 1.9 or 2.2 times higher than that of levelDB. The amount of writing is reduced by 60 and 65.
【作者单位】：清华大学计算机科学与技术系;
【基金】：国家自然科学基金项目(61327902,61433008) 北京市科委课题(D151100000815003)~~
【分类号】：TP311.13;TP333
【正文快照】： This work was supported by the National Natural Science Foundation of China(61327902,61433008)and Beijing MunicipalScience and Technology Commission of China(D151100000815003).(qinxiongjun2010@163.com)基于键值对(key-value)的key-value数据库是一种非关系型

【相似文献】