Building, Using and Managing the Data Warehouse

Building a data warehouse is based in the traditions of decision support, data modeling, and information center computing. Yet it is not reducible to any of these and supersedes them. Drawing on practices, technologies, and challenges that were not dreamt of until the mid 1990s. As George Zagelow's introductory essay makes clear, the data warehouse is a transformation of operational data into a from that provides business information, intelligence, and knowledge. His recommendation is to drive out nconsistencies in legacy data spanning data marts (smaller departmental assemblies). The enterprise- wide warehouse is built as the UNION of data marts where company-wide questions can be answered. If it works, this is a tactic for constructing the larger warehouse department by department. It does presuppose acceptance (or imposition) of an enterprise perspective. Mark Sweiger provides a thorough technical briefing for management on the role of massive parallel processing (MPP) as a method for tackling the large amount of data stored in warehouses. Technology is not a silver bullet here. But it is an important component of the answer. The author lays out the terms of the debate of the function shipping model (move the query to the data) versus dynamic data redistribution (rebalance skewed data by moving the data). These are not really on a collision course; and savvy administrators will look for a relational optimizer that can cost out the differences in access method. Next essays by Paul Barth and Robert Small / Herb Edelstein address examples, issues and tools in data mining applications. The first level data warehouse answers questions about which customers are buying which products or services and where and when they are doing so. But what if you either don't know what to ask or want to drill down as to why by chasing statistically significant correlation with demographic data. What if help is needed in formulating hypotheses? Then the advance to data mining is in order. Unlike the relational database model, which, in its many implementations is arguably an open standard, data mining tools are still exclusively proprietary (vendors specific lock-in is implied). Decision trees, neural networks, clustering and class analysis are the order of the day. Here the embedded technical function are pattern matching, bottom up rather drilling down through aggregates, and fuzzy logic rather than symbolic. If you recall the advertisement criticizing the consultant for quoting Sun Tzu on THE ART OF WAR, but being missing in action at implementation time, then you will want to study Bernard Boar on Understanding Data Warehousing Strategically. Without the strategic dimension, the use of the warehouse is without vision; and Boar provides that in good measure, including the references to Sun Tzu. According to Boar, the data warehouse provides the basis for a "rising tide" strategy. In this case, al

I found the book to be a bit dated in some technical areas (DW on a whole changes rapidly), but I really like the fact that the book is a conglomeration of MANY expert opinions. Most books on DW are from a single point of view, while this one combines the knowledge of many since it is from The Data Warehousing Institute. I found the areas on Managing and Staffing particularly helpful.