A BI solution describes the decision-orientated collection and processing of data for presenting relevant business information for the management. They are built upon several layers, or respectively components like ETL-tools (extraction, transformation, and loading), data warehouse (DWH), reporting, online analytical processing (OLAP) and data mining. Further, a consistent BI solution has to conform with certain business and technical requirements of an enterprise, which are based on data warehousing [CBS02, page 909]:

Data integration: A BI solution integrates heterogeneous, intra and external cooperated data sources into a homogeneous environment. Furthermore, it is important to define data consistently to offer users a uniform view of integrated data sources.

Time-referenced data: Data are stored over a long period of time, because they are only valid and accurate for a specific point and period of time. Data in a BI solution are past-referenced and previous data, which represent only a ”snap-shot” [CBS02, page 909].

Data constancy: It is necessary to refresh data sets from operative data systems (data resources), because data in a DWH are not updated in real time. Data sets in a BI solution are never replaced but they are supplemented with new data. Such new data sets are continuously integrated with previous data in a BI solution [CBS02, page 909.Based on this fact a BI solution provides both a granular and an aggregated data view.

Subject-orientated data: A BI solution is arranged according to a company's key processes (e.g. customers, products and sales) instead of application areas (issuing an invoice, storage or distribution) [CBS02, page 909]. Based on these requirements a BI solution provides decision-orientated accurate data sets in respond to information queries and to support decisions very quickly.

On the basis of these requirements, a conceptional BI solutions are built upon a five-tier architecture as shown in figure 1 and embraces different aspects across a company. Each tier can be represented by a separate BI software component depending on their specific processing requirements such as data warehousing or reporting.

Figure 1: Five-tier architecture of a classical BI platform.

The Data Source Layer includes the available or maybe multiple data sources of a business company such as ERP systems, DBMSs or legacy systems as well as flat files. A ETL (extraction, transformation, and loading) tool in the Input Layer describes which way and what kind of transformation occurs when data items move on from their sources, through the ETL process [Cog02, page 29] to the Storage Layer or target DWH. In essence, the ETL process cleanses and consolidates the multiple data sources into a uniform organisation view. Furthermore, the input layer contains already the overall metadata-driven architecture for data warehousing and all further business analysis in the Presentation Layer. In general, the DWH in the Storage Layer represents the consolidation and aggregation of the corporation-wide and distributed data sources. It stores the extracted and transformed data of the data sources. Furthermore, it contains only historical data and thus the DWH ”is designed for query and analysis rather than for transaction processing” [ORA03, Data Warehousing Concepts]. Regarding this, a DWH is exclusively updated in regular time intervals by the ETL process and not directly by end users. The Output Layer is represented typically by multiple data marts which are contain derived data subsets of the overall DWH “for a particular line of business” [ORA03, Data Warehousing Concepts]. A data mart feeds the front-end tools with data in the Presentation Layer. Within the Presentation Layer occurs the real business decision making with corresponding reporting and analysis tools or concepts. For the implementation of a BI solution, every layer needs metadata in the form of data warehouse objects and data models. Based on this, it is necessary to define appropriate data maps for the data workflow and processing in each BI solution layer.

Download article: BI_Architecture.pdf