Mapping problems within heterogeneous distributed data base systems

Abstract

Supporting heterogeneous database systems in a distributed database environment requires the translation between different forms of data representation and of data manipulation while transmitting data within the network. A generally accepted approach to this translation is to define a standard database system interface (data model and data manipulation language) and to build a global schema of the distributed data. If there is a need to move data from one network node to another, the request for data is first translated into the standard interface, and then converted into the destination database interface to be evaluated. The general problem of translations between different database interfaces is known as the "mapping problem". In this paper, we consider the mapping problem in a heterogeneous distributed database environment.First, two components of database mappings are considered: the data mapping that expresses the source data structure's elements in terms of the target data structure's elements, and the operator mapping that translates the source operators into the corresponding target operators using as reference the data mapping.Four different types of mapping are identified according to the relationship between the source and the target data strucutes. First, the trivial mapping where both the source and the target data structure express exactly the same data structure using the same data model. In this case, the data mapping, as well as the operator mapping, is totally expressed by an identity function. Second, the schema mapping where the source and target data structure differ, even though the underlying data model is the same. Third, the model mapping where the data structures expressed in both schemas are exactly the same, but are described using different data models. Finally, the total mapping occurs when both source and target data models and data structures differ. Each of these four mapping cases presents its own inherent difficulties that are analyzed through the paper.Then mappings are considered within Distributed Data Base Systems. We propose an architectural arrangement that localizes total and schema mappings between the user and the standard interfaces, and limits mappings between the standard and local interfaces to the trivial and model mapping cases. Several advantages result from this arrangement (which does not lessen the mapping flexibility between user and local interfaces). The mappings between the standard and local interfaces are certified feasible and in their simplest form, the design and evaluation of the standard schema is facilitated and the operator decomposition and scheduling are rendered independent of schema mappings.Lastly, the paper presents the architectural framework and the mapping capabilities of a heterogeneous distributed database system being implemented at the University of Southwestern Louisiana.