Abstract:
Integrating data from multiple, heterogeneous databases and other information sources has been one of the leading issues in database research and industry. There are two approaches towards solving the data integration problem--Multidatabase Systems and Data Warehousing. This thesis contributes towards solving the problem of data integration using the data warehousing approach. This thesis argues that the operational data store (ODS) fails to provide true operational integration and introduces a new data integration architecture by defining an architectural construct--the Enterprise Data Store (EDS). An Enterprise Data Store is a repository of data that represents an integrated view of enterprise operations and is built for corporate-wide operational informational processing and transactional processing of common business operations. This thesis presents an architecture and a comprehensive set of algorithms for synchronizing the EDS with the operational systems. The philosophy behind the EDS synchronization architecture is to exploit the metadata component of the data warehouse system. A very important component of the data warehouse metadata store is the mapping between the operational systems and the data warehouse. This research, based on this component of the data warehouse metadata store, identifies four kinds of mappings--entity to entity, attribute to attribute, key to key, and record to record mappings that can be used to synchronize the EDS with the operational systems. These mappings are modeled in a metadata model which is implemented as the metadata. mapper. The mapping data and algorithms stored in the metadata mapper are then used by the synchronization algorithms to synchronize the EDS with the operational systems. The proposed synchronization architecture offers many advantages and is different from early synchronization architectures (e.g., WHIPS) that are based on a materialized view approach.
