Normalization is the process of reducing or completely eliminating the occurrence of redundant data in the database. Normalization refers to the process of designing relational database tables from the ER model. It is a part of the logical design process and is a requirement for online transaction processing (OLTP) databases. This is important because it eliminates (or reduces as much as possible) redundant data. During the normalization process, you usually split large tables with many columns into one or more smaller tables with a smaller number of columns. The main advantage of normalization is to promote data consistency between tables and data accuracy by reducing the redundant information that is stored. In essence, data only needs to be changed in one place if an occurrence of the data is stored only once.
The disadvantage of normalization is that it produces many tables with a relatively small number of columns. These columns have to then be joined together in order for the data to be retrieved. Normalization could affect the performance of a database drastically. In fact, the more the database is normalized, the more the performance will suffer.
Traditional definitions of normalization refer to the process of modifying database tables to adhere to accepted normal forms. Normal forms are the rules of normalization. They are a way to measure the levels or depth that a database is normalized to. There are five different normal forms; however, most database solutions are implemented with the third normal form (3NF). Both the forth normal form (4NF) and the fifth normal form (5NF) are rarely used and, hence, are not discussed in this chapter. Each normal form builds from the previous. For example, the second normal form (2NF) cannot begin before the first normal form (1NF) is completed.
Note
A detailed discussion of all of the normal forms is outside the scope of this book. For help with this, refer to the Wikipedia article at http://en.wikipedia.org/wiki/Database_normalization.
In 1NF, you divide the base data into logical units called entities or tables. When you design each entity or table, you assign the primary key to it, which uniquely identifies each record inside the table. You create a separate table for each set of related attributes. There can be only one value for each attribute or column heading. The 1NF eliminates the repetition of groups by putting each one in a separate table and connecting them with a one-to-many relationship.
The objective of 2NF is to avoid the duplication of data between tables. In 2NF, you take data that is partly dependent on the primary key and enter it into another table. The entity is in 2NF when it meets all of the requirements of 1NF and has no composite primary key. In 2NF, you cannot subdivide the primary key into separate logical entities. You can, however, eliminate functional dependencies on partial keys by putting those fields in a separate table from the ones that are dependent on the whole key.
The 3NF objective is used to remove the data in a table that is not dependant on the primary key. In 3NF, no non-key column can depend on another non-key column, so all of the data applies specifically to the table entity. The entity is in 3NF when it meets all of the requirements of 1NF and 2NF and there is no transitive functional dependency.
Denormalization is the reverse of the normalization process, where you combine smaller tables that contain related attributes. Applications such as online analytical processing (OLAP) applications are good candidates for denormalized data. This is because all of the necessary data is in one place, and SQL Server does not require to combine data when queried.