Denormalization

Denormalization is the deliberate reversal of normalization: combining or duplicating data that a normalized design would keep in separate tables, so that a query can retrieve what it needs with fewer joins, at the cost of storing the same fact in more than one place.

A dimension table flattening a product's category, subcategory and brand into one row, rather than splitting them into three linked tables, is a denormalization; so is a one big table design that pre-joins facts and every relevant dimension together. The trade-off is direct: fewer joins and faster reads, in exchange for more storage and the risk that a repeated value becomes inconsistent if it is updated in one place but not another.

Denormalization is the default posture of most analytical modeling, including the star schema and dimension table approach, because analytical workloads are overwhelmingly read-heavy and joins are typically the most expensive part of a query at scale; it is used far more cautiously in transactional systems, where frequent writes make redundancy and the risk of inconsistent copies more costly. A materialized view is effectively a controlled, automated form of denormalization, since it stores a pre-joined result without requiring the base tables themselves to be restructured.