Research On Flash-Bsed Database Storage Structure

Flash memory, a new storage medium, is gradually replace the dominant position of magnetic disk.Due to historical reasons, the page storage structure of relational database is mainly designed for the characteristics of the disk, which could not make full use of flash memory. At present research on flash memory database is an important issue, but is mainly focus on the optimization of the traditional page storage structure. The performance of flash memory is still restricted by the traditional architecture.This paper presents a new type of storage structure which mainly aims at the imbalanced speed of reading and writing characteristics of flash memory. The data has been divided into Blocks which are the smallest write unit with the same size of the erase unit of NAND flash memory and Pages which are the smallest read unit, while one Block containing many Pages.In order to further optimize the access performance, this paper proposes a new compression algorithm based on the new database storage structure. The compression algorithm needs the entire Block to compress, while only need a small piece of data which belong to one Page to decompress, which ensure the good compression rate and need not read entire compressed data to find the small piece of data which we want, which not only reduce the IO, but also reduce the CPU overhead of the decompression.Through the experiment, application of FBSS database storage structure greatly reduce the data size, and bring good random access performance, which is of great significance to improve the database IO bottleneck.