An Introduction To Dimensional Modeling For Data Warehousing - Part 1

The design principles of the dimensional model, whichhand, maintaining aggregate facts, limits the analysis
is commonly used in data warehousing, are described(or drill down) capability on certain dimensions. The
in this article series. Dimensional models capturemodel in the figure (see resource), captures an
business performance measurements, which are usedevent: the sale of a product at a given time and all
to support decision making. Dimensional model The(or most) related to the sales event dimensions. This
descriptive simplicity and high performance in queryfact table type, is called a transaction fact table.
execution, are characteristics which have contributedDimension tables Dimension tables describe the
to the increased use of the dimensional model in datadimensions of a measurement on a business process.
warehouse infrastructures. The symmetry andThe features of each dimension should be as rich and
descriptive simplicity can be seen at the conceptualflexibly described as possible (with many descriptive
model (see resource link) which relates to retail salesfields on the dimension table). Attribute names of the
monitoring (data warehousing technology has beendimension tables should be sufficiently descriptive, so
introduced initially in retailing).as to be easily and unambiguously understood.
Relational data models are use to implement theCodes which are used in operational systems, should
above conceptual model (as depicted in the resourcebe replaced with descriptive names of the
link).characteristics. Numerical (quantitative) measurements
This model is easily understood by Business analysts,should not be entered in dimension tables (given that
in contrast with other operational systems modelsthese facts should be stored in fact tables). Non
('normalized data models' in relational modelingnumerical measurements (e.g. measurements which
language). In a relational modeling representation, thecan be described in text) which are derived from a
model in its simple form, consists of a central 'factlist of discrete values, should be entered in a
table' and 'dimension tables' which are connected todimension table. The dimension tables usually maintain
the 'fact table' via reference keys (foreign keys ina limited number of records (the different
relational modeling language). This form is called 'stardescriptions that a dimensional entity may take) (the
schema'. Fact table The 'fact table' is the central tablenumber of rows is known as the cardinality of the
in a dimensional model, which stores thetable). The attributes of dimension tables, play an
measurements (facts) on which analytical processingimportant role in dimensional analytical processing,
takes place. All measurements adhere to the samegiven that they form the base of all 'restriction
level of detail. The most useful measurements areoperations' which are applied. (e.g. sales that took
usually additive or semi-additive, in order to allowplace on a specific branch and date are derived by
analytical processing (numerical calculations whichrestricting on the branch and the date dimension).
produce additional 'derived' facts). During analyticalMoreover, they form the headings in the reports
processing, thousands or millions of fact rows areproduced. Therefore, dimensional attributes are the
retrieved and numerical processing is applied on facts'entry points' to the measurements which are
or fact combinations. Facts which are stored in a factcaptured in the fact table. The value of a dimensional
table, should be captured at the most detailed (ormodel is directly proportional to the quality and depth
most granular) level (also called 'atomic level', meaningof its dimension tables.
something that cannot be divided). On the other