Database tables and indexes may be stored on disk in one of a number of forms, including ordered/unordered flat files, ISAM, heap files, hash buckets, or B+ trees. Each form has its own particular advantages and disadvantages. The most commonly used forms are B+ trees and ISAM. Such forms or structures are one aspect of the overall schema used by a database engine to store information.
Unordered storage typically stores the records in the order they are inserted. Such storage offers good insertion efficiency (), but inefficient retrieval times (). Typically these retrieval times are better, however, as most databases use indexes on the primary keys, resulting in retrieval times of or for keys that are the same as the database row offsets within the storage system.
Ordered storage typically stores the records in order and may have to rearrange or increase the file size when a new record is inserted, resulting in lower insertion efficiency. However, ordered storage provides more efficient retrieval as the records are pre-sorted, resulting in a complexity of .
Most conventional relational databases use "row-oriented" storage, meaning that all data associated with a given row is stored together. By contrast, column-oriented DBMS store all data from a given column together in order to more quickly serve data warehouse-style queries. Correlation databases are similar to row-based databases, but apply a layer of indirection to map multiple instances of the same value to the same numerical identifier.