Database Reference
In-Depth Information
For industrial database work, the traditional approach for high level data modeling is
to use a version of Entity Relationship (ER) modeling (Chen, 1976), such as the Information
Engineering (IE) approach (Finkelstein, 1998), the Barker version of ER modeling (Barker,
1990), or IDEF1X (Integration Defi nition 1 extended). Although the original 1993 version
of IDEF1X has a standard metamodel (NIST, 1993), we ignore it here since it is actually a
hybrid of ER and relational modeling, and its successor, IDEF1X
97
, also known as IDEFobject
(IEEE, 1999), has so far been largely ignored by the marketplace.
More recently, Unifi ed Modeling Language (UML) class diagrams (OMG, 2003) and
the Object-Role Modeling (ORM) approach (Halpin, 2001a) have also gained popularity for
information modeling. Following its adoption by the Object Management Group (OMG),
the UML is now the de-facto standard in industry for object-oriented code design. ORM is
a fact-oriented approach that can be used as a conceptual front-end to attribute-based ap-
proaches such as ER and UML, and is currently being considered by the OMG's Business
Rules Special Interest Group as a candidate for business rule modeling at the computation-
independent level.
A modeling language can be specifi ed by a
metaschema
, which is a schema that indicates
the grammatical structures to which any application schema formulated in the modeling
language must conform. Strictly, a
model
is the union of a schema (structure) and a popula-
model
tion of instances (e.g., objects or facts that instantiate the information-bearing structures in
the schema). A metaschema supplemented by structures to capture specifi c populations is a
metamodel
. In practice, the term “metamodel” is sometimes loosely used as a synonym for
“metaschema”. While published metamodels for UML (OMG, 2001, 2003) have been widely
debated, and many suggestions have been made to improve UML (e.g., see Siau & Halpin,
2001), it is diffi cult to fi nd any in-depth analysis of metaschemas for the other approaches.
This paper provides new metaschemas for two ER approaches (IE and Barker) as well as
ORM to reveal their commonalities and differences, and to address modeling issues such
as the use of derived associations and the virtues of orthogonality. UML has been examined
previously (e.g., Halpin & Bloesch, 1999; Halpin, 2001b) and is quite complex; hence only
an incomplete analysis of its metamodel for data modeling is given here. For a detailed
comparative evaluation of all the methods, including IDEF1X, see Halpin (2001a).
The next section of this chapter provides a metaschema and related discussion of the
IE notation. The two sections after that metamodel the Barker ER and ORM approaches,
respectively. We then evaluate the different approaches to multiplicity in UML, ER and
ORM. Some other aspects of the UML metamodel are then discussed. The fi nal section
summarizes the main contributions, notes some advantages of an attribute-free modeling
approach, and lists references for further reading.
model
is the union of a schema (structure) and a popula-
INFORMATION ENGINEERING
The
Information Engineering
approach was originated mainly by Clive Finkelstein,
Information Engineering
who developed a modeling procedure for the notation and extended IE to Enterprise En-
gineering (EE). Finkelstein (1998) provides an overview of IE with further details on his
website (www.ies.aust.com/~ieinfo/). The IE notation was later adapted by Martin (1993).
Although Martin's recent topics favor the UML notation, IE is still used far more exten-
sively for database design than UML, which is mostly used for object-oriented code design.
Different versions of IE exist, with no single standard. In one form or another, IE has long
