Databases Reference
In-Depth Information
A significant amount of research has recently been dedicated to automatic
map-
ping adaptation
(
Yu and Popa 2005
) to support schema evolution and is surveyed in
chapter 7 (
Fagin et al. 2011
). This work mostly assumed relational or nested rela-
tional schemas and different kinds of logical schema mappings. For these settings,
the definition and implementation of two key operators, composition and inversion
of mapping, have been studied. These operators are among those proposed in the
context of model management, a general framework to manipulate schemas and
mappings using high-level operators to simplify schema management tasks such as
schema evolution (
Bernstein 2003
;
Bernstein and Melnik 2007
). A main advantage
of composition and inversion is that they permit the reuse of existing mappings and
their adaptation after a schema evolves. The proposed approaches for mapping adap-
tation still have practical limitations with respect to a uniform mapping language,
mapping functionality, and performance so that more research is needed before their
broader usability.
3.3
Summary
Tab le
6.1
shows a side-by-side comparison of most of the approaches described in
this section for key requirements of Sect.
2
. With the exception of the Panta Rhei
project, all solutions focus on the simple (table) changes of SQL DDL. Oracle is the
only system that also allows the specification of changes by providing a new version
of a table to be changed as well as a column mapping. Commercial GUIs exist that
can support simple diffing and change bundling, but eventually output simple SQL
DDL without version mappings or other versioning support. Oracle's edition con-
cept makes versioning less painful to emulate, though underlying physical structures
are still not versioned. Overall, commercial DBMS support only simple schema
changes and incur a high manual effort to adapt dependent schemas and to ensure
backward compatibility. PRISM adds value by enabling versioning through inter-
version mappings, forward and backward compatibility, and formal guarantees of
information preservation when applicable. HECATAEUS improves flexibility by
specifying how to update dependent schema objects in a system when underlying
objects evolve.
4
XML Schema Evolution
XML as a data model is vastly different than the relational model. Relations are
highly structured, where schema is an intrinsic component of the model and an
integral component in storage. On the other hand, XML is regarded as a
semi-
structured
model. Instances of XML need not conform to any schema, and must
only conform to certain well-formed-ness properties, such as each start element
having an end tag, attributes having locally distinct names, etc. Individual elements
