理解C-Store about column-oriented database

ikeepo

于 2021-03-31 18:15:40 发布

阅读量478

点赞数 2

分类专栏： Database 文章标签： C-Store column-oriented

本文链接：https://blog.csdn.net/The_Time_Runner/article/details/115356992

版权

Database 专栏收录该内容

25 篇文章

订阅专栏

C-Store 使用列式存储（ROS）来高效管理数据，其中每个列文件按特定属性排序并压缩。新数据存储在未压缩的写优化存储（WOS）中，定期通过后台进程转移到ROS。投影是按相同属性排序的列组，通常至少有一个包含所有可用于查询的列。C-Store 不支持传统索引，但提供稀疏索引以快速访问列的物理页。更新操作被视为删除后插入，删除通过特殊‘删除列’记录。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

Overview

In C-Store, the primary representation of data on disk is as a set of column files which is called “read optimized store (ROS)”.

Each column-file contains data from one column, compressed using a column-specific compression mothod, and sorted accroding to some attribute in the table that the column belongs to.

Newly loaded data is stored in a write-optimized store (WOS) where data is uncompressed and not vertically partitioned.

Periodically, data is moved from the WOS into the ROS via a background “tuple mover” process, which sorts, compresses, and writes re-organized data to disk in a columnar form.
Projections

Groups of columns sorted on the same attribute are referred to as “projections”;

Typically there is at least one projection containing all columns that can be used to answer any query;
Index

C-Store does not support secondary indices on tables

A sparse index is a small tree-based index that stores the first value contained on each physical page of a column.

理解physical page in Computer
No-overwrite

C-Store uses a “no-overwrite” storage representation, where updates are treated as deletes followed by inserts.

Deletes are processed by storing a special “delete column” that records the time every tuple was deleted.