Categories Software

Continued from Introduction

The type of data warehouse an organization adopts should depend on the way the business operates and the types of decision support it needs.

One of the simplest types of data warehouse, an operational data store (ODS) is a replicated production database that has been adjusted for errors. An ODS is used primarily to generate standard operations reports and to provide transaction detail for summary-level analysis. (Since an ODS replicates an OLTP system, some experts don't consider it a true data warehouse type. We've included it in our discussion, however, because ODSs fit the broad definition and many data warehouses contain them.)

Depending on an organization's reporting needs, an ODS may be updated monthly, weekly, or more frequently, sometimes almost in real time. Its main advantage is that it enhances production system performance, since reporting and query functions are off-loaded from the OLTP system to the ODS.

If your organization just generates operational reports and does little analysis or marketing research, an ODS may well suit your needs. On the other hand, if your company handles relatively few transactions per day, an ODS may be overkill. You can use the production system to generate reports instead.

Another type of data warehouse is the data mart. Data marts are limited in scope, usually taking their information from a single department or business process. They may be used for analyzing sales information in a specific region or for a particular product line, for example. Data marts usually contain only summary data, but they can be linked to operational data stores for drilling down to transaction details if necessary. They are sometimes managed by IT departments, but just as often they are managed directly by users in a department or work group.

While many OLAP applications can be performed on data marts, cross-departmental analysis, executive information systems, and data-mining applications need information gathered from the entire enterprise to be most effective. The enterprise data warehouse is used for this type of extensive data collection and analysis. Because of its scope and complexity, the enterprise data warehouse is usually managed by the central IT group.

As its name implies, an enterprise data warehouse contains information taken from throughout an organization. This is the most complex type of warehouse to build and maintain, since data must be merged from multiple systems into common subject areas.

Different systems often produce incompatible or inconsistent data, and sometimes the data must undergo several transformations before it can be integrated into a data warehouse in a meaningful way. Inmon estimated that 80 percent of the time, building such a data warehouse would be spent on extracting, cleaning, and loading data.

If your organization can benefit from data mining, building an enterprise data warehouse may be well worth the effort. Data-mining tools work with various statistical techniques for modeling data and for estimating and predicting outcomes based on what they have learned. They work best with large data sets.

There are many ways to build an enterprise data warehouse. One form uses a centralized approach, combining all the enterprise's data in a single, large data store. Another form takes a distributed approach, pulling data from multiple data marts.

Next: Data Warehouse Components

Published as Enterprise Computing in the 3/9/99 issue of PC Magazine.

Related Links
• ZDNet Enterprise