What is Normalization?

Last updated Mar 28, 2023 - Edit Source

• Data engineering

Normalization is used in relational database design to reduce data redundancy and improve data integrity. Developed by British computer scientist Edgar F. Codd in the 1970s as part of his relational model, normalization involves organizing the columns (attributes) and tables (relations) in a database to ensure proper enforcement of dependencies through database integrity constraints.

This is achieved by applying formal rules during the synthesis (creation of a new database design) or decomposition (improvement of an existing database design) process.

1. First Normal Form (1NF):
   • Eliminate duplicate data by ensuring each attribute contains only atomic values and each table has a unique primary key.
2. Second Normal Form (2NF):
   • Meet all requirements of 1NF and remove partial dependencies by ensuring that every non-prime attribute (attribute not part of any candidate key) entirely depends on the primary key.
3. Third Normal Form (3NF):
   • Meet all requirements of 2NF and remove transitive dependencies by ensuring that no non-prime attribute is transitively dependent on the primary key.

Denormalization, on the other hand, is the process of intentionally introducing redundancy into a database design by combining tables or adding redundant data, aiming to improve query performance or simplify the database structure. Denormalization is the opposite of normalization. Please consider the trade-offs between data integrity and query performance. This technique is used with Dimensional Modeling in OLAP cubes, for example.