闪存数据库顺序和随机操作性能深度测试和分析

摘要

Flash memory has begun to be more widely used by both the individuals and the enterprises as an available novel storage medium for the alternative of magnetic disks.Although the solid state disks (SSDs) significantly outperform the magnetic disks in I/O speed,the performance of flash-based database systems has not been speeded up correspondingly yet due to lack of consideration of the features of flash memory.Detailed testing and studying on the difference between sequential and random operations performance of SSDs in this paper further proves that modem SSDs has preferably faster random access speed.We then choose the representative database benchmarks and adjust the query plans of various queries to observe the performances of database system in different query plans.We find that some of the query plans can better capture the random access characteristics of SSDs,and one of them runs even 6 times faster than the default query plans.Finally,we summarise the shortcoming of the encounter of traditional database system against the medium of SSD and put forward some suggestions in regard to optimisation.%闪存作为一种可供替代磁盘的新型存储介质,已经开始越来越广泛地被个人和企业使用.固态硬盘(SSD)的速度大大超出磁盘,但因为数据库系统并未完全发挥闪存的特性,闪存数据库的性能并没有得到相应的提升.详细测试和研究SSD的顺序和随机操作的差异,进一步证明现代SSD具有较快的随机访问速度.选取具有代表性的数据库基准,调整各查询的查询计划,观察不同查询计划下数据库系统的性能.由此发现,某些查询计划可以更好地发挥出SSD的随机访问特性,其中某个查询的最佳性能甚至比使用默认的查询计划快6倍以上.最后,总结传统数据库系统面对SSD这一介质的短板,并提出一些优化的建议.