Article Preview
Top1. Introduction
Enterprises throughout the world use business intelligence processes, technologies and tools to transform enormous amount of data from operational systems into existing, actionable and decision ready information. This information supports the organization’s decision-making process and helps in making future strategies that drives lucrative commercial action (Sauter, 2010; Turban et al., 2005). Different business processes gather and handle data in loosely coupled or independent form and thus, the data becomes inconsistent and the information obtained using it could be misleading (Inmon, 2003). The best possible solution to this problem is a well-designed data warehouse, which forms a central repository of integrated data and is used for transforming data into information (Kimball and Ross, 2002; Mohania et al. 1998). Data warehouse, in addition to being integrated, contains subject-specific and time variant non-volatile data with the goal to support business decision making (Inmon, 2003). Thus, data warehouse is key to business intelligence. Decision support applications access data warehouse using online analytical processing (OLAP) tools to extract information. The analytical queries involves summarization and aggregation of data from the continuously growing data warehouse These analytical queries, in order extract relevant and required information from data in data warehouse, take long time for processing (Chaudhuri and Dayal, 1997; Gupta et al., 1997, Theodoratos and Sellis, 1997). Several query optimization techniques exist that minimize the query response time in order to expedite the decision making process. Amongst these, view materialization and indexing has been found to be effective (Harinarayan et al., 1996) and is also the focus of this paper. In order to aid and support analytical processing, Star schema storage representation is used by a data warehouse to store data. Star schema comprises a single fact table, which stores business measures, surrounded by several dimension tables, which describe the measures and are related to the fact table through a primary-foreign key relationship. The result of analytical queries, which join facts and some of the dimensions is limited by restrictions enforced in the dimensions. Since the size of fact table is enormous, joining tables is a very time-consuming operation (Chan et al., 1999; Chirkova and Yang, 2011). Join operations can be optimized by implementing table views, which are created by joining dimension tables in accordance with the OLAP query. However, it is not an efficient approach in the context of a Decision Support System (Chan et al., 1999; Chirkova and Yang, 2011), as enormous quantities of data is processed and alike queries are recurrently repeated in warehousing applications. A virtual view does not store data but fetches data in accordance to the query posed on it. Therefore, for efficient data analysis, the views are pre-aggregated and pre-computed on the relevant dimensions and stored in the data warehouse. Such views are referred to as materialized views. Materialized views drastically improve the query performance by offering fast lookups on the precomputed data. Answering queries using materialized views takes less response time, as they are lesser in size in comparison to a data warehouse (Harinarayan et al., 1996). Selecting a suitable set of views that minimize the total query response time and the maintenance cost while conforming to the limited storage space is key to efficient and effective data warehouse design (Agrawal, 2000; Chirkova et al., 2001; Nadeau and Teorey, 2002). This is termed view selection, which is discussed next.