基于SSD的Key-Value系统优化设计与实现

发布时间：2019-02-16 21:11

【摘要】：随着互联网技术的迅猛发展,越来越多的非结构化数据涌入到人们的生活中,为这些数据建立高效的索引面临极大的挑战.键值数据库Key-Value以其结构简单和高扩展性而引起人们的广泛关注,已成为海量数据存储系统中的重要组成部分.由于Key-Value系统对吞吐量要求较高,而基于Flash的固态硬盘(solid state drive,SSD)能够提供很高的随机读性能,在SSD上构建Key-Value系统已成为海量数据存储领域的一大研究热点.鉴于Flash具有非定点更新、寿命有限等特性,基于SSD的KeyValue系统必须针对Flash的特性作专门优化.以一种称为SkimpyStash的基于SSD的Key-Value系统为基础,提出了一种新的Key-Value系统低延迟存储系统(low latency store,LLStore).LLStore使用内存文件映射技术来减少针对SSD的IO请求,除此之外,针对SkimpyStash中低效的压缩策略,提出一种改进方法,可以在少量增加内存开销的情况下极大地减少查询时间.通过与原系统的性能比较实验,LLStore在平均查询时间上可以获得至少12%的加速.
[Abstract]:With the rapid development of Internet technology, more and more unstructured data are coming into people's lives. It is a great challenge to build an efficient index for these data. Key and value database (Key-Value) has become an important part of mass data storage system because of its simple structure and high scalability. Because Key-Value system requires high throughput, and solid state hard disk (solid state drive,SSD) based on Flash can provide high random read performance, constructing Key-Value system on SSD has become a research hotspot in the field of mass data storage. In view of the non-fixed update and limited life of Flash, the KeyValue system based on SSD must be specially optimized for the characteristics of Flash. Based on a Key-Value system based on SSD called SkimpyStash, this paper proposes a new low latency Key-Value storage system, (low latency store,LLStore). LLStore, which uses memory file mapping technology to reduce IO requests for SSD. Aiming at the inefficient compression strategy in SkimpyStash, an improved method is proposed, which can greatly reduce the query time with a small increase in memory overhead. By comparing the performance of LLStore with that of the original system, the average query time of LLStore can be accelerated by at least 12%.
【作者单位】：高性能计算国家重点实验室(国防科学技术大学);国防科学技术大学计算机学院;
【基金】：国家“八六三”高技术研究发展计划基金项目(2012AA012402)
【分类号】：TP333

【参考文献】