Knowledge and Data Management in GRIDs

1. Introduction (p. 20)

The Grid, as an emerging infrastructure for the discovery, access and use of distributed computational resources [15], offers new opportunities and raises new challenges in data management. Many aspects differentiate the Grid from a traditional distributed environment, such aspects include the large scale, dynamic, autonomous, and distributed nature of data sources.

A Grid can include related data resources maintained in different syntaxes, managed by different software systems, and accessible through different protocols and interfaces. Due to this diversity in data resources, one of the most demanding issue in managing data on Grids is reconciliation of data heterogeneity [8].

Therefore, in order to provide facilities for addressing requests over multiple heterogeneous data sources, it is necessary to provide data integration models and mechanisms.

Data integration is one of the most persistent problems that the database and information management community has to deal with. Although significant progress has been made in several aspects of data integration, the increase in availability of web-based data sources has led to new challenges. More specif- ically, efficient techniques have been developed and approaches have been devised to schema mediation languages, query answering algorithms, optimisation strategies, query execution policies, industrial development, and so on [17].

However, effective techniques for the generation and handling of semantic mappings are still in their infancy. The need for semantic correlation of data sources is particularly felt in Grid settings. Moreoever, in a Grid, a centralized structure for coordinating all the nodes may not be practical because it can be- come a bottleneck and, more importantly, it cannot accommodate the dynamic and distributed nature of Grid resources.

Data access and integration services have been attracting significant interest from the Grid community. Data Grids that rely on the coordinated sharing of and interaction across multiple autonomous database management systems play a key role in many industrial and scientific initiatives. To this end, middleware services have been developed.

Two notable examples are the OGSA Data Access and Integration (OGSA-DAI) [6] and the OGSA Distributed Query Processor (OGSA-DQP)' [5,4] projects. These projects have moved toward a servide-oriented architecture quite early in their lifecycle. OGSA-DAI exposes database management systems (including Oracle, MySQL, SQLServer, DB2, and so on) in a uniform way, whereas OGSA-DQP provides distributed query processing functionalities on top of OGSA-DAI. As such, OGSA-DQP can combine and integrate data from multiple data sources. To enhance performance, it employs parallel query execution techniques, nevertheless it relies on the user for the semantic interpretation of the data and does not address any schema integration requirements.

To date, only few projects (e.g., [ l l , 91) actually meet the schemaintegration requirements that are necessary for establishing semantic connections among heterogeneous data sources. To address this limitation, the use of the XMAP framework for integrating heterogeneous data sources distributed over a Grid has been proposed [12] .