Semantic heterogeneity in multidatabase systems: a review and a proposed meta-data structure.

ABSTRACT

This article reviews current research on the issue of semantic conflict resolution in multidatabase system design. It is observed that in multidatabase systems, semantic conflicts need to be resolved at both schema level and instance level. Based on the literature review, a new taxonomy for differentiating semantic conflicts and a meta-data representation incorporating the taxonomy are proposed. It is argued that the new meta-data representation is effective for summarizing local schemata, and hence it can serve as a common protocol for multidatabase systems that require instance level conflict resolution.

Keywords: meta-data representation; multidatabase; semantic conflict resolution; semantic heterogeneity.

INTRODUCTION

Semantic heterogeneity or semantic conflict is the main source of problems in multidatabase design. In this article, a brief review of previous work in semantic conflict identification is presented which leads to the creation of a taxonomy for resolving conflicts in multidatabase design that is more inclusive when compared to existing frameworks, for example that of Batini et al. (1986). A meta-data structure, based on this taxonomy, will be proposed that can be used as a point of reference (a common protocol) for semantic conflict resolution.

For last three decades, a significant quantity of multidatabase research has focused on resolving the problem of semantic heterogeneity or semantic conflicts. Semantic heterogeneity is often present in multidatabase systems because of the lack of global schema definition. The situation is similar to common misunderstandings that occur in everyday interpersonal communication. Misunderstandings can result from two people who speak different languages. They cannot understand one another unless interpreters are present. Even when interpreters are used, concepts that cannot be precisely translated remain. In fact, the level of shared understanding between the parties after communication depends heavily on the knowledge of the interpreters. Even if persons participating in the conversation are speaking the same language, misunderstandings could persist due to the ambiguity of language. Based on this analogy, it is apparent that not all semantic conflicts can be systematically resolved. A good conflict resolution system should have the intelligence to separate resolvable and irresolvable conflicts. Given such an intelligent conflict resolution process, corresponding ]procedures can be created to integrate results from different data sources and to report inconsistencies that cannot be resolved.

To create multidatabases without semantic conflicts, a significant amount of research has focused on schema integration at the conceptual level (Lim & Chiang, 2000). Real world examples such as the Cyc knowledge base in Carnot (Collet, Huhns & Shen, 1991; Singh et al., 1997) and the CORDS multidatabase (Martin & Powley, 1997) all use similar schema-integration concepts to provide multidatabase systems with an integrated view at the logical level. However, in practice, semantic conflicts exist not only at the logical or conceptual level, but also at the instance or run-time level. That is, in practice, many conflict resolutions may need to be performed at query run-time. To facilitate run-time semantic conflict resolution, the integration engine should have the ability to construct consistent meta-data at run-time. In this article, a meta-data structure for the purpose of capturing the run-time meta-data generated by such a run-time integration problem is proposed. We organize this article by addressing the following questions in sequence:

* What is a multidatabase system?

* What are the methods currently used in multidatabase systems to resolve semantic heterogeneity?

* What are different types of semantic heterogeneity?

* Can a "better" taxonomy for classifying semantic heterogeneity be found, resulting in a meta-data structure to assist in addressing semantic conflicts?

* Is the meta-data structure proposed sufficient for practice?

WHAT IS A MULTIDATABASE SYSTEM?

Database systems are a major resource in most corporations today. Organizations have been buying state-of-the-art software platforms on which database systems are built. Typically, a database system is designed to address an organization's needs at a fixed point in time. Organizations, however, have information needs that are dynamic. The original design of a particular database system could soon be and often is quickly outdated. This situation is inevitable in fast growing or multi-site organizations where different subunits have developed their own database systems. When the greater organization requires integrated data, problems emerge because database systems cannot "talk" or query one another directly. A mediator is required to coordinate the communication and/or data exchange process. The design and implementation of such a mediator is an essential component of so-called multidatabase systems.

A multidatabase system provides integrated access to heterogeneous, autonomous local databases (Bright, Hurson & Pakzad, 1994). It resides unobtrusively on top of existing database systems and presents the illusion of a single database to its users (Dogac, Dengi & Oszu, 1998). Even though the concept of multidatabase systems has been around for several decades, the development of multidatabase systems still faces many obstacles. One obstacle is the semantic heterogeneity or semantic conflict between different database systems. The main source of these conflicts is the different data abstraction/representation mechanism used by different designers. Even worse is the presence of the many data modeling methodologies that exist in today's computing environment. Database design methodologies have evolved from those based on hierarchical models to the currently dominant relational database design. Object oriented database models have now appeared on the commercial database market that will result in further refinement of the design method. However, not every organization changes its database system whenever a "better" system is available. In fact, many large organizations are still using database systems that are more than 25 years old (Miller, Yu & Nilakanta, 2002). Database systems that are based on different designs increase the complexity of the schema integration problem.

Although there is no consensus, the term multidatabase system usually refers to a distributed information-sharing system. Unlike the term distributed-database, a multidatabase system always implies the integration of heterogeneous database systems. In the literature, multidatabase systems are also called federated database systems, multidatabase language systems and interoperable systems. The basic requirements/assumptions for a multidatabase system include the following (Bright, Hurson & Pakzad, 1992; Grandi, 1998):

1. The local DBMSs have independent meta-data and exist before joining the multidatabase system.

2. The local DBMS should participate in the multidatabase system with little or no modification.

3. The local DBMS retains autonomy with full control over local data and processes.

Lim and Chiang (2000) articulate the concept of coupling the local DBMSs to build a multidatabase by introducing two important dimensions of database integration: schema versus instance and physical versus virtual. Schema integration refers to the integration of meta-data at the time the multidatabase is designed, while …