Abstract
Information systems were developed in early 1960s to process orders, billings, inventory controls, payrolls, and accounts payables. Soon information systems research began. Harry Stern started the “Information Systems in Management Science” column in Management Science journal to provide a forum for discussion beyond just research papers (Banker & Kauffman, 2004). Ackoff (1967) led the earliest research on management information systems for decision-making purposes and published it in Management Science. Gorry and Scott Morton (1971) first used the term ‘decision support systems’ (DSS) in a paper and constructed a framework for improving management information systems. The topics on information systems and DSS research diversifies. One of the major topics has been on how to get systems design right. As an active component of DSS, which is part of today’s business intelligence systems, data warehousing became one of the most important developments in the information systems field during the mid-to-late 1990s. Since business environment has become more global, competitive, complex, and volatile, customer relationship management (CRM) and e-commerce initiatives are creating requirements for large, integrated data repositories and advanced analytical capabilities. By using a data warehouse, companies can make decisions about customer-specific strategies such as customer profiling, customer segmentation, and crossselling analysis (Cunningham et al., 2006). Thus how to design and develop a data warehouse have become important issues for information systems designers and developers. This paper presents some of the currently discussed development and design methodologies in data warehousing, such as the multidimensional model vs. relational ER model, CIF vs. multidimensional methodologies, data-driven vs. metric-driven approaches, top-down vs. bottom-up design approaches, data partitioning and parallel processing.
TopBackground
Data warehouse design is a lengthy, time-consuming, and costly process. Any wrongly calculated step can lead to a failure. Therefore, researchers have placed important efforts to the study of design and development related issues and methodologies.
Data modeling for a data warehouse is different from operational database data modeling. An operational system, e.g., online transaction processing (OLTP), is a system that is used to run a business in real time, based on current data. An OLTP system usually adopts Entity-relationship (ER) modeling and application-oriented database design (Han & Kamber, 2006). An information system, like a data warehouse, is designed to support decision making based on historical point-in-time and prediction data for complex queries or data mining applications (Hoffer, et al., 2007). A data warehouse schema is viewed as a dimensional model (Ahmad et al., 2004, Han & Kamber, 2006; Levene & Loizou, 2003). It typically adopts either a star or snowflake schema and a subject-oriented database design (Han & Kamber, 2006). The schema design is the most critical to the design of a data warehouse.
Many approaches and methodologies have been proposed in the design and development of data warehouses. Two major data warehouse design methodologies have been paid more attention. Inmon et al. (2000) proposed the Corporate Information Factory (CIF) architecture. This architecture, in the design of the atomic-level data marts, uses denormalized entity-relationship diagram (ERD) schema. Kimball (1996, 1997) proposed multidimensional (MD) architecture. This architecture uses star schema at atomic-level data marts. Which architecture should an enterprise follow? Is one better than the other?