By Peggy King
Participants in the plenary session on data warehousing on Friday afternoon of UniForum '96 concluded that this application set is no longer just a pipe dream. Yet neither does it offer established solutions.
Rebel Brown, president of the consulting firm Cognoscenti, moderated a panel consisting of executives from two vendor companies and two users who are implementing large-scale, second-generation data warehousing projects. Brown addressed some of the issues that make data warehousing difficult. "IT professionals are faced with the challenge of combining disparate information models and with making sure that their users are looking at the right data," she said.
Vendor panelists Chris Erickson, president of Red Brick Systems, and Steve Sommer, vice president of marketing for Informix Software, gave overviews on how data warehousing has evolved from large-scale decision support to a distinct type of business application. While both men agreed that a database optimized for online transaction processing cannot deliver the performance desired for ad hoc queries, they took opposing views, predictably tied to each of their firm's type of technology, about whether a specialized database is needed for this task. Erickson, whose company sells a database tailored for data warehousing, delivered a word of caution to audience members planning to implement such a project. "A company needs to establish an entire information architecture designed around its own data access needs before it considers building a data warehouse," he said.
Sommer discussed some of the types of data warehouses his company's customers have built. He argued that a "universal" database can be tuned and optimized to be a data warehouse as long as it has high-availability features, the ability to handle mixed data types and good performance for multidimensional analysis queries.
Both of these early implementers faced technical and business challenges in scaling their first-generation data warehouses to the capacity and performance levels that their organizations require. For Makishima's organization, the principal goals for the new data warehouse architecture were to migrate to Unix servers from HP's corporate mainframe and to increase the timeliness of the information made available to users within the WCSO. The sheer size of the implementation--with tables as large as two gigabytes refreshed on a monthly basis--was a major challenge.
IMS America built its existing data factory three years ago with a bank of Unix workstations to deliver prescription information by zip codes. This proved inadequate for a market that increasingly required information delivered on a per-patient level of granularity. In order to reach this next level, the company needed to add three terabytes of data storage. "We looked at all the existing products and found that our requirements overwhelmed all of them," said Brownstone.
All panelists agreed that data warehousing is a steep hill to climb. Erickson advised customers to find a qualified systems integrator to help with the initial implementation, but he warned that some of the larger firms are just getting up to speed with the technology and may bring inexperienced staff to the project. Brownstone urged healthy skepticism toward vendor claims, recommending that users adopt the Missouri attitude: "Show me." Both customer panelists agreed that, despite all the data access and analysis tools now available, producing better reports is still the single benefit that end users appreciate the most.