From Conventional to Multiversion Data Warehouse - Evolving Application Domains of Data Warehousing and Mining - page 44

maintain meta-information related to evo-
lution. Due to a limited metadata structure,
it is not possible to store information about
all those evolution operations that result in
data warehouse evolution.

to note that the concepts of simulating business
scenarios and classification of versions (Bebel,
2004) (to real and alternative versions of the data
warehouse), are ignored to: a) keep our focus on
versioning of data warehouse and b) propagate the
concept of multiversion data warehouse clearly
and unambiguously.
Before describing multiversion data warehouse
as a solution to the problems discussed in the previ-
ous section, basic concepts related to dimensional
modeling are discussed briefly next.

Conventionally, data warehouses rarely have
the ability to handle source schema changes
(Nasir, 2007; Samos, 1998), because they are
not designed for that purpose. It is possible that
changes in sources result in specific changes in
contents of the data warehouse that cannot be
handled (Golfarelli, 2004). Changes in the data
warehouses include insertion/deletion/updating
of fact, dimension or dimension-level (Balaschka,
1999). To be version-compatible, the metadata
structure change operations need to be recorded
and different versions need to be accessible in
change-prepared data warehouses. Therefore,
metadata structure should be extended to store a
set of operations that could produce changes in the
data warehouse without loss of any information.
Additionally, the new solution should be easily
acceptable, adoptable and usable for existing data
warehouse users. For this last reason, it should
be possible to interact with versions of the data
warehouse using conventional SQL-like query
languages with few or no extensions.

Basic concepts: Star Schema

The logical structure of a data warehouse is called
a dimensional schema, which is used to store
subject-oriented and time-variant data for deci-
sion support. The dimensional schema has four
types: star, snowflake, constellation and star flake
schema (Kimball, 2002). Out of these types, star
schema is the central concept and the most often
used element (Ponniah, 2001). The star schema is
labeled so due to its star-like structure, as shown
in Figure 1.
A dimensional model (DM) is composed of fact
table(s) and a set of dimension tables (Kimball,
2002). a) Fact table consists of two types of attri-
butes, called measures and key attributes. Facts or
measures are numeric measurements to represent
a critical value for an enterprise (Paulraj, 2001).

verSIonIng A dAtA WArehouSe

Figure 1. Example star schema

In the remaining part of this chapter, the multi-
version data warehouse is described as a solution
to the problems discussed above. The principles
of creation and management of versions in data
warehouse and framework for version creation
are presented in this section.
A data warehouse that is composed of multiple
schema and instance versions is called multi-
version data warehouse (MVDW). In MVDW, a
schema version is the structure of a data warehouse,
and an instance version is the dataset described by
a schema version (Wrembel, 2005). It is important