The T-ADAPT site of Dr. Michael Hahne

T-ADAPT

Introduction

Temporal aspects in Data Warehousing

Conceptual design with ADAPT

Extension from ADAPT to T-ADAPT

Summary

References

Model Management

Model Transformation

SAP BW Models

SQL Server Models

Visio Shapes

Hahne online

Imprint

T-ADAPT:

Conceptual design of temporal aspects in data warehousing with T-ADAPT

from Dr. Michael Hahne

Chapter 3: Conceptual design with ADAPT

Different approaches in literature for the conceptional modelling of multidimensional information systems were discussed [GG98; Sc00]. Apart from modifications of entity relationship models and object-oriented approaches also suggestions with an own methodology arise. The Application Design for Processing Technologies (ADAPT) spread in America represents such a method which was introduced by Bulos in 1996 [Bu96]. ADAPT is a modelling approach which explicitly focuses multidimensional data structures for OLAP applications.

In Germany ADAPT has only been discussed for some years [Bu98; discusses more in detail in TJ98]. The emphasis of the modelling with ADAPT lies on the representation of dimension structures and hierarchical structures. Their representation is discussed in section 3.1. Section 3.2 briefly deals with aspects of modelling cubes.

It is referred to a constant example for the elucidation of the connections in the following. Its subject is an enterprise of the electronics industry dealing with production and selling of electronic devices and components. Besides relevant electronic construction units as for instance resistances, transistors, condensers etc. also laboratory instruments and products of the High Fidelity area belong to the program. The distribution of products is made by mail order business on basis of a catalogue as well as by Internet on basis of an E-Commerce solution.

3.1: Modelling dimensions

The strong point is the representation of various dimension structures and their economical reference. The individual components for modelling dimensions and also the defined hierarchical structures on them are summarized in tables 6 and 7.

	Dimension (Axis of a cube)
	Hierarchy (path of aggregation)
	Level in a hierarchy (consolidation level)
	Dimension attribute (additional information to dimension members)
	Dimension member (uniquely defined value)
	Dimension scope (subset of a dimension)
	Model (formula for calculating derived data)

Table 6. Objects for modelling dimensions in ADAPT

Besides the actual objects also the connection among those is important. A summary of connecting types for dimension modelling is given in table 7.

	Loose precedence
	Strict precedence
	Self-precedence

Table 7. Objects used for modelling relationships between levels

The representation of substantial components which are used for representing dimensions in ADAPT is based on the example of a partially shown product dimension in figure 3. The consolidation in the product hierarchy is done by brands, product categories and product sub-categories. An aggregation to the different manufacturers of the products is considered beyond that. Product caption, packing size and type as well as weight are also further relevant information besides the actual product number. They are attached to the ADAPT model as attribute of the product level. Self production and external procurement of the products are differentiated, whereby the object of the dimension scope is used.

Fig. 3. Example of a product dimension represented in ADAPT

One or more hierarchies are the feature of a dimension following multidimensional basic understanding. These are based on super ordination and subordination conditions and form the consolidation paths in dimensions. They are thereby the basis for operations of drill down and roll up. It can be derived from the representation that two hierarchies are defined for products which are differently developed in each case.

The representation shows a dimension defined by levels, so that all hierarchies are modelled on dimension levels. These different hierarchies are often called parallel hierarchies. The common basis level is represented by the object for the product level by which the finest granularity in this dimension is specified.

The vendor hierarchy is a first simple case since only an aggregation from products to vendors is done. Exactly one vendor is assigned to each product.

The used symbol with the double arrow for the connection of the levels clarifies the strict relationship; therefore exactly one super ordinate element is assigned to each product. In the other hierarchy in which the level?s brand, category and subcategory are differentiated this is not the case. However, there can also be products which are assigned to no subcategory but are directly connected to a category. The notation of this connection is done in ADAPT with the connecting arrow with only one arrow.

Attributes to dimensions or dimension levels are another main object for the representation of multidimensional structures on the semantic level. The attributes are exclusively on the lowest level of all products in the illustrated example and model further characteristics of products as for instance their weight as well as further packing-relevant information.

It has often to be decided for various economical objects whether they should be modelled as attribute or as an own dimension in the modelling phase.

ADAPT?s strength is the modelling of dimension scopes representing a logically connected subset of a dimension. For example, subsets for self producing and external procurement are conceivable for products. These dimension scopes are related to each other. For instance a product is either produced or bought additionally. This type of relation is called fully exclusive. Table 8 shows an overview of the different relation types between dimension scopes.

	Fully exclusive subset
	Fully overlapping subset
	Partially exclusive subset
	Partially overlapping subset

Table 8. Relationship types in ADAPT

Exclusive subset concerning the designation of relations means that the subsets involved are disjoint. However, the partially overlapping subset permits laps. The combination of the subsets forms the whole, it is called fully. Otherwise it is described as partial.

Dimensions of direct development with the type of object for dimension elements are also possible besides level determent dimension types. Typically, these are dimensions for scenario and measures [element-defined type of dimension in Gl01]. In fig. 4 the example for the consideration of different selling ways of a dimension is shown.

Fig. 4. Dimension defined by elements

The dimensions firmly put the structure of the multidimensional data in cubes. These are the subject of the following section.

3.2 Modelling of cubes

The cube object is available for the representation of a cube in ADAPT. The caption and the connected dimensions are registered manually in them. The linking of dimensions involved is visualized by a connecting arrow object. In general modelling by the Cube object and the attached dimension objects are enough in order to reduce the complexity of representation. However, the complete representation is more meaningful concerning smaller models since all model components are visible at once. Exemplarily the customer marketing Cube in which the dimensions are represented directly is modelled in figure 5.

Analysis with respect to temporal aspects is recognized to offer also information about the financial years besides the standard selections of years, quarters and months in the example since these deviate in an enterprise from calendar years. The representation in ADAPT is done with two separated hierarchies in the dimension time. Special days as for example working-days and the last days of a month which are modelled as dimension scopes in the example are as important as vacation days for analysis. The holidays are modelled as attribute of the calendar date. Also seasonal aspects are to be considered and can be modelled well with dimension scopes.

Fig. 5. Representation of a cube in ADAPT

Special attention of evaluations is on the accumulated information about customers concerning direct customer business by catalogue or on-line activities. Those information is grouped regionally like represented in the example. Further a grouping according to family status, sex and age can be done. Additionally, detailed information such as last name, first name, regional information, etc. has to be made available to each customer apart from the customer number. The demographic attributes are defined as dimension scopes. Name and address information is modelled as attributes.