Picking Data from the Warehouse

Analysts often associate OLAP with data warehouses, those large repositories in which businesses collect data for an entire enterprise (which may include historical data spanning a number of years for all of a company's divisions). For day-to-day operations, a business needs only relatively current data, and each operating unit needs only its own data.

But large-scale analysis, for which OLAP is designed, generally needs a historical record over time, encompassing all of a company's operating units and product lines for comparison purposes. So, even though OLAP need not imply large databases -- the multidimensional analysis produced by OLA P pro ducts can be useful for small businesses as well -- in practice, businesses often consider OLAP as part of their data-warehousing strategy.

However, OLAP analysis generally requires summary data, not a complete duplication of the warehouse-detail data. For example, if a customer buys a certain product several times during a certain time period, the OLAP database needs only the totals, not the record of each individual transaction. Moreover, an OLAP database need not contain information irrelevant to trend analysis and forecasting, such as addresses or contacts.

So, an OLAP database normally contains only a portion of the information in the data warehouse on which it is based. Because of the way multidimensional databases structure their data, identifying information (e.g., product name, time period, and so forth) resides only in the dimension-header area (comparable to the row and column headers of a spreadsheet) rather than being repeated for each record, as in a relational database. The net effect is a multidimensional database requiring considerably less disk space than a data warehouse.