A physical data model introduces the database-specific context missing in conceptual and logical data models. It represents the tables, columns, data types, views, constraints, indices and procedures within the database and/or the information communicated during computer processes.
Physical data models should be built in relation to a specific database management system (DBMS) as well as the specific requirements of the processes that operate based on the data. This often requires denormalization of logical design constructs to maintain referential integrity. An example of the contextual considerations at the physical data modeling stage is the nature of the data that can/will be processed and the rules regarding how such processes can be executed.
Another key consideration is ensuring the modeled column types are supported in the DBMS, and the naming conventions for entities and columns are observed, preventing problematic semantic overlaps. The consideration of technological context means physical data models reflect the needs of the technological environment as is, or as intended.
The three types of data models can and should be viewed as linear stages. As the third stage of data modeling, physical data modeling builds on the models developed in the conceptual and logical stages.
Physical models mark a shift in models being primarily constructed to represent “what” – as in the data and information that will be modeled – to “how” – implementation. Naturally, this practical approach to implementation takes into account the specifics of the DBMS and technology, including denormalization requirements, proposed for the project.
The model describes a single project’s data needs, but it also can be integrated with physical data models from other projects to account for the interrelationships between projects, processes and technology. Due to its considerations of specific technology, a physical data model is more rigid, and even small changes can require modifications to the entire application. Therefore, it’s advisable to only progress to constructing a physical data model when the conceptual and logical data models have been built.