Abstract: In large databases, empty fields in high dimensional sparse table may occupy a large amount of storage space. To deal with this problem, a slicing based database structure, which simulates the principle of column-store database in traditional row-store database, is designed. By analyzing on testing results, the storage space is decreased about 27.42% lower than the original model. The number of I/O data block is reduced to 35.27% when five fields of high-dimensional sparse data are queried, and 28.22% when four fields are selected. There will be a further reduction when the number of fields reduces. Therefore the accessing efficiency for database is improved.

Key words: high-dimensional, sparse data, column-store database

摘要： 针对大型数据库中高维稀疏关系表空字段对存储空间的占用问题，通过利用传统行存储数据库模拟列式存储数据库的工作原理，设计了一种基于分片的数据库结构。通过实验分析，数据库存储空间比原始模式降低了27.42%左右。在对高维稀疏数据中五个字段进行查询时，I/O数据块个数降低至原始模式的35.27%，对高维稀疏数据中四个字段进行查询时，I/O数据块个数降低至原始模式的28.22%，而随着字段的减少I/O数据块仍会进一步减少，从而提高了数据库的访问效率。

关键词: 高维, 稀疏数据, 列式存储数据库