Relationship Database Design

Content

• Conceptual model - ERD
• Logical model - schema, data dictionary, attributes mapping, ...
• Physical model

E-R Diagram (Conceptual Model)

Superclass/Subclass Hierarchy

"O" is overlapping, meaning there could be overlapping between each subclass entities. (A manager could also be a salesperson)
Besides, we can use "D" for disjoint subclasses.

A single line between Employee and its subclasses implies an optinal participation (an employee may not belong to any subclass), while a double line implies a mandatory participation (an employee must be in one or more subclasses).

An U-like symbol identify subclasses by pointing to them. (the entity the bottom of "U" points to is a subclass)

Schema represents abstract (strong and weak) entities and associated attributes and necessary relationships between entities.

A schema looks like this:
STUDENT( Stu_num, Stu_Lname, Stu_Fname, program_num@, age, gender)
(PK "Stu_num" should be underscored. I use italic instead)

Data dictionary contains metadata of attributes, with which you could understand the what, where, how, how much, who, when questions about data.

A good logical model should follow normal form of database.

• UNF
  Contains repeating values.

• 1NF
  Each cell in the table contains only one value (so that the table can be represented in a relational database).

• 2NF
  Satisfies 1NF;
  No partial dependencies.
  Meaning: A non-primary-key attribute cannot be dependent on part of primary-key.

For example, in R( A B C D), if
(A+B) -> C, D,
(B) -> D,
D is partially dependent on primary key (A+B), thus not satisfying 2NF.

• 3NF
  Satisfies 2NF;
  No transitive dependencies.
  Meaning: A non-primary-key attribute cannot determine non-primary-key attributes. (But a non-primary-key attribute can determine a primary-key attribute)

• BCNF
  Satisfies 3NF;
  No non-CK dependencies.
  Another way to say this: For all functional dependencies exist in the table, all their determinants are candidate keys.
  Meaning: the second and third normal forms normalize non-prime attributes, and BCNF normalizes relationships between prime attributes.

For example (a model in 3NF but not in BCNF)
In R( A B C D), there are
(A+B) -> C, D;
(A+C) -> B, D;
(C) -> B.
With a non-primary-key attribute C determines primary-key attribute B, the model does not satisfy BCNF.

• 4NF
  Satisfies BCNF;
  No multi-value attributes.