Independent Data Marts: Each data mart is developed independently, each with its own dimension and fact tables. Data from various source systems are transferred to each data mart separately; each source system may feed one or more data marts. Users query each data mart separately for needed information. Data marts may represent data inconsistently, and it is difficult or impossible to combine data from different data marts.
Data Mart Bus Architecture: Data marts use common dimensions and definitions and are linked together via middleware to maintain consistency. Data extracted from multiple source systems can be shared across multiple data marts. Users can query one or more data marts for needed information, and data from multiple data marts can be combined by the middleware to respond to a query. Performance may be slow especially for complex queries. Maintaining consistency across all the data marts can be a challenge.
Hub-and-Spoke Architecture: Data from multiple source systems are transferred into a single enterprise data warehouse (EDW), which in turn feeds several data marts for different subject areas. Each user queries the appropriate data mart for his or her subject areas of interest. This arrangement ensures consistency because all data come from the common EDW while allowing reports and user interfaces to be customized for each user group through the individual data marts. There may be data redundancy because the same data are stored in multiple locations (in the EDW and in one or more data marts). Transferring data to the EDW and then to the data marts before making it available to users may create delays, so users are not seeing the most up-to-date data.
Centralized Data Warehouse: Data are transferred from sources to a single enterprise data warehouse (EDW). Users query the EDW directly for information without going through data marts. This simplifies data management and administration, reduces the need to store multiple copies of the data, and minimizes delays so users can get more up-to-date data. It may be more difficult to customize reports and user interfaces for different groups of users.
Federated Architecture: Instead of a physical data warehouse and data marts, source systems are connected together through middleware. Instead of transferring data, user queries are fulfilled directly from the original data in the source systems. Middleware creates a logical view of the source systems to present users with what appears to be a single unified data source, even though it is made up of disparate physical systems behind the scenes. Transferring data into a data warehouse or data marts and maintaining separate copies of the data are eliminated. Because of issues with data quality and slow performance, this approach is currently used mainly to supplement rather than replace the data warehouse.
[End of Transcript]