The end goal of database design is to be able to transform a logical data model into an actual physical database. A logical data model is required before you can even begin to design a physical database. Assuming that the logical data model is complete, though, what must be done to implement a physical database?
The first step is to create an initial physical data model by transforming the logical data model into a physical implementation based on an understanding of the DBMS to be used for deployment. To successfully create a physical database, you must have a good working knowledge of the features of the DBMS including:
- In-depth knowledge of the database objects supported by the DBMS and the physical structures and files required to support those objects.
- An understanding of how the DBMS supports indexing, referential integrity, constraints, data types, and other features that augment the functionality of database objects.
- Detailed knowledge of new and obsolete features for the specific versions or releases of the DBMS to be used.
- Knowledge of the DBMS configuration parameters that are in place and whether any may need to be tweaked to support the new database.
- Data definition language (DDL) skills to translate the physical design into actual database objects.
Armed with the correct information, you can create an effective and efficient database from a logical data model. The first step in transforming a logical data model into a physical model is to perform a simple translation from logical terms to physical objects. Of course, this simple transformation will not result in a complete and correct physical database design—it is simply the first step. The transformation consists of the following:
- Transforming entities into tables
- Transforming attributes into columns
- Transforming domains into data types and constraints
To support the mapping of attributes to table columns, you will need to map each logical domain of the attribute to a physical data type and perhaps additional constraints. In a physical database, each column must be assigned a data type. Certain data types require a maximum length to be specified. For example, a character data type could be specified as CHAR(25), indicating that up to 25 characters can be stored for the column. You may need to apply a length to other data types as well, such as graphic, floating point, and decimal (which require a length and scale) types.
Keep in mind that no commercial DBMS supports relational domains. Therefore, the domain assigned in the logical data model must be mapped to a data type supported by the DBMS. You may need to adjust the data type based on the DBMS you use. For example, what data type and length will be used for monetary values if no built-in currency data type exists? Many of the major DBMS products support user-defined data types, so you might want to consider creating a data type to support the logical domain, if no built-in data type is acceptable.
In addition to a data type and length, you also may need to apply a constraint to the column. Consider a domain of integers between 1 and 10 inclusive. Simply assigning the physical column to an integer data type is insufficient to match the domain. A constraint must be added to restrict the values that can be stored for the column to the specified range, 1 through 10. Without a constraint, negative numbers, zero, and values greater than ten could be stored. Using check constraints, you can place limits on the data values that can be stored in a column or set of columns.
Specification of a primary key is an integral part of the physical design of entities and attributes. A primary key should be assigned for every entity in the logical data model. You should try to use the primary key as selected in the logical data model. However, multiple candidate keys often are uncovered during the data modeling process. You may decide to choose a primary key other than the one selected during logical design—either one of the candidate keys or another surrogate key for physical implementation. But even if the DBMS does not mandate a primary key for each table, it is a good practice to identify a primary key for each physical table you create. Failure to do so will make processing the data in that table more difficult.
Of course, there are many other decisions that must be made during the transition from logical to physical. For example, each of the following must be addressed:
- The nullability of each column in each table
- For character columns, should fixed length or variable length be used?
- For many columns, you may have to choose between multiple similar data types (e.g., SMALLINT, INTEGER, and BIGINT). Does the data model provide sufficient details to choose properly?
- Should the DBMS be used to assign values to sequences or identity columns?
- Implementing logical relationships by assigning referential constraints.
- Building indexes on columns to improve query performance.
- Choosing the type of index to create (if multiple types are available to you): b-tree, bit map, reverse key, hash, partitioning, etc.
- Deciding on the clustering sequence for the data.
- Other physical aspects such as column ordering, buffer pool specification, data files, denormalization, and so on.
And we have not even begun to consider aspects such as recovery time objects for making sufficient backups, the scheduling of reorganization, and so on.
A logical data model should be used as the blueprint for designing and creating a physical database. But the physical database cannot be created properly with a simple logical to physical mapping. Many physical design decisions need to be made by the DBA before implementing physical database structures. This may necessitate deviating from the logical data model. But such deviation should occur only based on in-depth knowledge of the DBMS and the physical environment in which the database will exist. And, of course, every deviation should be properly documented for posterity.