Enterprise Data Modeling Practices - High-Performance Web Databases: Design, Development, and Deployment - page 163

DATA ADMINISTRATION: QUALITY ASSURANCE AND SELF-CONTROL
Quality cannot be reached simply by defining formal objectives for Qual-
ity Assurance. In some cases explicit goals for quality assurance will help
lead activities down the right path. This is also true for data modeling activ-
ities. In the absence of other goals, some Data Administration departments
adhere to normal forms and similar theoretical approaches as indicators of
quality — but this is not the essential goal. It is better to use business-driven
objectives such as integration, reuse, and modeling of the core business.
Quality Objectives
There should be a documented statement of the Data administration
department's quality goals. The goals will be either business driven or the-
ory driven — the order of goals will imply which.
Reuse
Reuse is an explicit goal of many IS organizations and should be mea-
sured. The reuse quota is an indicator for the quality of data modeling
activities.
Typical reuse quotas from projects are about 50 percent. Some are less
and some are more, depending on the project and previous activities.
Projects retrieve information about known entities from a variety of
sources. For example, information can be retrieved from a host-based
repository only (an adequate method) or from discussions with the Data
administration group during the design process (a better method) as well
as from a repository.
Data Integration at the Physical Level
Physical data integration across several applications is an indicator for
good data modeling practice. Another good sign is the presence of logical
objects that are implemented over and over again.
Reference models are not off-the-shelf production data models. They
have to be fitted to fulfill the individual needs of the organization using
them. They should best be separated physically or logically from the pro-
duction data model.
The Data administration group may want to analyze its data integration
attempts by posing the following questions:
• Is there a physical border between logical, physical, and reference
data models?
• Is there a clear logical border? Is it possible to see at first glance
whether an entity is in production or an entity from a reference model?
• Is the reference model subject to a version concept?