The architecture of Data Warehouse systems is described on basis of so-called reference architectures. Today's requirements to Enterprise Data Warehouses are often too complex to be satisfactorily achieved by the rather rough descriptions of this reference architecture. We describe an architecture of dedicated layers to face those complex requirements, and point out additional expenses and resulting advantages of our approach compared to the traditional one.
Characteristics of an Enterprise Data WarehouseA Business Data Warehouse (BDW, [1]) is a Data Warehouse (DW) to support decisions concerning the business on all organizational levels. It covers all business areas, such as logistics, finance, and controlling. We define Enterprise Data Warehouses (EDW) as subspecies of BDW systems, which not only cover all activities of an enterprise, but are an important basis for applications, such as Business Intelligence, planning, and Customer Relationship Management. An EDW collects and distributes huge amounts of data from a multitude of heterogeneous source systems. It has to provide a single version of truth for all data of the company. That means that there is one common view on centralized, accurate, harmonized, consistent, and integrated data at a given point of time. The range of use is often world-wide, so that data from different time-zones have to be integrated. Frequently, 24x7-hours data availability has to be guaranteed, facing the problem of loading and querying at the same time. In addition, there are high requirements to the data: ad-hoc access, near real-time availability, high data quality, and the need for very detailed and granular data with a long time horizon. Moreover, new or changing requirements for information have to be flexibly and promptly satisfied, and, last but not least, the data access has to be secured by a powerful authorization concept. These significant requirements for an EDW also necessitate enhancements and refinements to the DW architecture, compared to the traditional one, which is outlined in this paper. Sections 2 and 3 describe the traditional and the layered DW architecture and show a comparative example of both approaches. In Section 4, we discuss additional expenses and resulting advantages of the layered architecture approach in detail. Section 5 concludes our paper and gives an outlook on future work.
