Databases Reference
In-Depth Information
The first of these is similar to Hull's notion of maintaining information capac-
ity [
Hull 1984
]; the second is one of the requirements in
Batini et al.
[
1986
].
Based on these requirements, the outcome of the merge is a set of relations, where
each of the relations of the first three types of constraints (i.e., all but containment
constraints) mean that the relations are combined, with all of their attributes present
in the global schema. For containment constraints, the two relations are left separate,
since there can be no way of determining what relational algebra operator to use.
3.2
Casanova and Vidal
Casanova and Vidal
[
1983
] describe a method for performing view integration. In
particular, they break the problem into two parts: combination and optimization.
The combination phase consists of gathering all of the views, as well as deter-
mining the dependencies between schemas. The optimization phase concentrates
on both minimizing the schema and reducing redundancy. They concentrate on the
optimization phase, which is akin to the problem considered in semantic merge.
The kinds of dependencies that they consider are, as in
Biskup and Convent
[
1986
], instance based. They consider the following dependency types:
Functional dependencies
An inclusion dependency says that one relation contains a subset of the tuples in
another
An exclusion dependency says that the instances of a relation are disjoint.
A union functional dependency essentially allows functional dependencies to be
declared valid across different schemas
The authors show that in general trying to optimize an integration with the above
dependencies is intractable, so they concentrate on a limited subset of schemas.
Essentially, these restrictions limit where the different kinds of dependencies can be
applied, as well as assuming that the schemas are already in Boyce Codd Normal
Form (BCNF).
Their goal is to create an optimization that is minimal in size, and also removes
redundancy. They also want to ensure that any consistent state of the initial schema
can be mapped into a consistent state of the transformed schema and vice versa -
a notion very similar to the idea of completeness. To achieve these goals, their
transformation removes many inclusion dependencies or union functional depen-
dencies (since they may be a source of redundancy), as well as vacuous exclusion
dependencies.
They provide an algorithm that will perform the optimization for the limited
cases. Note that their algorithm creates the unified schema, but does not create the
mapping from the unified schema to the initial views.
There are a number of other schema merging works in the view integration
domain around this period, including
Shoval and Zohn
[
1991
]and
Navathe and
Gadgil [
1982
]. Generally, these works build on approximately the same foundation:
define what it means to merge and what it means for there to be a conflict. Most
