Database Reference
In-Depth Information
of them are scientific ones, but two of them are
included in commercial products. In this section
some of these approaches are presented and com-
pared by their capabilities.
poral and non-temporal data sources. This work
is based on the assumption of a data warehouse
being only a view over the sources. It contains
temporal query language and a framework to cre-
ate and incrementally update temporal views over
the history of instances. As such a history may
not always be available in the sources, auxiliary
data is stored in the data warehouse to enable
self-maintaining views.
Approach of Kimball: Slowly
Changing Dimensions
Kimball (1996) was probably the first discovering
the need for evolving structures in a data warehouse
and introduces three methods for dealing with
so called
slowly changing dimensions.
The first
method simply suggests overwriting old instances
with their new values, thus tracking a change his-
tory is not possible. The second method consists
of creating a new instance for each change. This
will create a version history, but means additional
effort in data management. One has to introduce
a surrogate key, because the natural primary keys
may not be unique any longer. For relating vari-
ous versions of an instance to each other, creating
a time stamp for the validity of each version is
proposed. The third method proposes creating a
new attribute for the instance, so the original and
the current attribute value can be saved. This can,
of course, only handle two versions of an instance.
All three methods are quite straightforward and
only allow very basic modifications on the instance
level. Impacts of structure changes on the cell
data are not considered. Furthermore, they are
only applicable for slowly changing dimensions
and not for what Kimball calls a
rapidly changing
monster dimension
(Kimball and Ross, 2002). For
this type of changes, they suggest swapping out
frequently changing attributes (e.g. age or income
of a customer) into so called
minidimensions.
With
their help the original member remains unchanged
and if an attribute changes, just another member
in the minidimension is used.
Approach of Hurtado,
Mendelzon and Vaisman
The approach of Hurtado, Mendelzon and Vaisman
(Hurtado, Mendelzon & Vaisman, 1999; Vais-
man & Mendelzon, 2001) proposes a temporal
multidimensional model and a temporal query
language. It allows modifications of schema and
instances. The temporal dimension schema is
defined as a directed acyclic graph. Each node
represents a dimension level. The edges connecting
the levels are labeled with a time interval denot-
ing the time when the edge is valid. The same
approach is used for dimension members, i.e.
the nodes represent dimension members and the
edges between them are labeled with their valid
time. This model assumes that only the relations
between the nodes, and not the nodes themselves,
change. The temporal query language TOLAP can
be used to execute queries over a set of temporal
dimensions and fact tables.
Approach of Blaschka, Sapia
and Höfling: Fiesta
The
Framework for Schema Evolution in Multidi-
mensional Databases (FIESTA)
(Blaschka, Sapia
and Höfling, 1999) supports a schema design
technique and some evolution operations. The
authors derive the need for an evolution methodol-
ogy from the fact that the data warehouse design
process is an iterative process. The proposed
evolution algebra supports modifications on the
schema level. Instance evolution is not supported
Approach of Yang and Widom
Yang and Widom (1998) present an approach that
allows building a temporal data warehouse for tem-
Search WWH ::
Custom Search