Column-oriented databases

From 2002 to 2006, I worked at a Canadian manufacturer of a column-oriented database.

Traditional databases store their data in individual records: record by record. This has advantages for transactional databases because the data needed in one operation are closely packed together. However, for analytical applications, one might be interested in, for example, the average consumption of all customers in a specific postal code area. Here, many individual records need to be considered, which can be “far” apart in memory. If the data are stored by their respective columns, this offers speed advantages. The individual columns can also be compressed more efficiently because the data are often very similar. Therefore, column-oriented databases are often better suited for analytical applications.

Some examples:

• MonetDB
• C-Store
• LucidDB
• Metakit