Data Warehouse Maintenance, Evolution and Versioning - Data Warehousing Design and Advanced Engineering Applications - page 183

Approach of Body, Miquel,
Bedard and Tchounikine

directly, but the instances are adapted automati-
cally according to the changes on schema level,
either physically in the database or logically with
views and filters. Body, Miquel, Bedard and Tchounikine (2003)

present their approach for handling evolution in
multidimensional structures. As basis for their
approach they define a temporal multidimensional
model.Based upon this model a set of evolution
operators is defined. The multidimensional model
comprises the typical elements of a data warehouse
structure, i. e. dimensions, levels, dimension
members and hierarchical relations. All of the
elements on the instance level are timestamped,
schema elements are not evolvable. Furthermore,
it contains a so called confidence factor, describing
the reliability of data, for instance if it is source
or mapped data, and a mapping relationship that
describes the mapping between two versions of a
dimension member. Operations are only defined
on the member level. There are operations for
creating new members, removing and reclassifying
(i. e. moving) members, and create associations
between versions of a member.

Approach of Quix

Quix (1999) provides a framework for quality-
oriented evolution of data warehouse views. The
author proposes a data warehouse process model
to capture dynamic aspects by representing the
data warehouse evolution as a special process. The
main focus is on the maintenance of evolving data
warehouse views. Metadata is provided to keep
track of the history of changes in the conceptual
and logical schema. Consistency rules are defined
guarantee consistency when quality factors have
to be re-evaluated.

Approach of Sarda

Sarda (1999) presents a formal model for a
temporal data warehouse that supports multiple
hierarchies, a symmetric treatment of dimensions
and measures and many-to-many relationships
among dimension levels. The authors provide a
mapping from their formal temporal ware house
model to a relational model. History is recorded
for dimension members as well as for relations
between them.

Approach of Kaas, Pedersen
and Rasmussen

Kaas, Pedersen and Rasmussen (2004) present
their approach for supporting schema evolution
for star and snowflake schema. In contrast to other
approaches, this methodology takes into account
the special needs of an evolving star or snowflake
schema using a relational database as data storage.
They provide a rich set of changing operations,
including operations for inserting and deleting
dimensions, categories and dimension members.
For each of the operations, the impact on existing
queries and the complexity of applying the operation
is evaluated. These evaluations are compared for the
case of an underlying star and snowflake schema.
From these comparisons the authors conclude that
a star schema is more robust in case of structure
evolution. Data transformation is not captured in
this paper, but considered as future work.

Approach of Ravat and Teste

Ravat and Teste (2000) introduce an object-
oriented approach to data warehouse modeling.
They define a Warehouse Class Extension (the
instances extracted from the sources) as current
state, a set of historical states, and a set of archived
states for the instances. Whereas historical states
are available on detailed level, archived states can
only be queried on an aggregated level. The main
focus is laid on modeling of data warehouses.