Browse Books

Reviewer: Lou Agosta

When data warehouse construction is driven by the need to understand customers, products, and key business events, data warehousing completes the promise of the client/server initiative. The promise is to provide access to data in a timely, flexible, and accurate manner. For those wondering what a data warehouse is, whether they might already be operating one, how to tell if one is needed, and how to build one, Kimball provides the answers. The fundamental distinction to be made in successful data warehouse construction is that between day-to-day operations and business strategy. The processes making up operations include highly granular transactions stored at the detail level, and online transactions processing (OLTP). This is the stuff of classic order entry, inventory control, and general ledger. Construction decisions require insight and vision about the performance and objectives of the business, based on facts about the business, delivered fast. Thus, data warehousing consists of modeling the business in terms of basic central facts (units sold or delivered, captured and summarized from operations) in relation to the fundamental dimensions that constitute the business over time. Typically, this results in a multidimensional model: a fact structure surrounded by product, customer (market), and time (history). This is the celebrated “star schema” discussed in the popular trade journals. Kimball claims his work is consistent with the online analytic processing (OLAP) movement, with one difference. His approach is “open,” employing de facto industry standard relational technology, whereas OLAP is still proprietary and, more important, not robust enough to scale to the enterprise level. Chapters 2 through 9 take the reader through a series of progressively more abstract examples of dimensional data modeling—in the grocery store, the warehouse, shipping, banking services, cable TV subscription services, and casualty insurance. A word of caution: if you are interested in a specific industry, you should not jump immediately to that chapter. It builds on groundwork laid earlier. For computing professionals, the wealth of functional business distinctions, especially in connection with the relational database model, is instructive. There are useful suggestions on how the relational model can be extended as well as on measures needed in application code until such extensions occur. For example, measures that record a static level, such as inventory and financial account balances, are not additive across time. Balances cannot simply be added, but must be averaged by time period. Since the SQL avg function considers rows returned, not time periods relevant ( periodavg ), an average period sum must be calculated in an application or by means of a proprietary SQL extension. The size of the performance challenge of data warehousing can be appreciated by considering product, customer, and time dimensions of an average of 10,000 distinctions each. Without sparsity (not all combinations occur), the result is a combination on the order of 100 billion rows. Naturally, the problem is made worse for phone companies and banks, which have millions of customers. (See chapter 6, “The Big Dimensions.”) Kimball claims that the limits of current relational technology, circa 1995, are reached at about 1 billion rows, or about 100 gigabytes. The answer is considered in chapter 13, on aggregation. Since an endless horizon of business days tends to cause combinatorial explosion of the facts at an elementary level, it is useful to define aggregations (summations) that group 20 or more facts together. Combine and store the data on a weekly or monthly basis, rather than daily. The tradeoff is between more work, transforming data in long-running batch processes prior to loading, and quicker online response time to queries submitted interactively. The book contains a wealth of practical advice for information technology practitioners. For example, when the relative size of the central fact table is compared with that of the surrounding dimension structures (differences of orders of magnitude are common), it is clear that little disk space can be saved in normalizing the latter. The book is superbly prepared. It includes a complete glossary of terms, appendices, an index—though no bibliography—and a useful summary of design principles of a dimensional data warehouse. It comes with a CD-ROM, which contains an ACCESS version of the databases described in the book and sample queries against the databases. The continuity with the disciplines of data modeling, data administration, and data mining is useful and productive. Much of what occurs in decomposing data into relational structure by means of Codd's normal forms is relevant here, but with a new spin. The structure of the data gives us insight into the kinds of questions that might be asked. Thus, the prospect of packaging a large but finite set of SQL queries can be envisioned. Kimball has taken the hype out of data warehousing and has shown its importance to business practices as an application of relational technology.

Reviewer: Alfs T. Berztiss

This excellent book describes established data warehouse technology—dependable data warehouses that can be built now. Far too often, writers try to combine coverage of established practice with the most recent research findings, which can leave readers confused. Although this author has outstanding research credentials, he has resisted the temptation to satisfy different reader interests at once. This book is not scholarly, so there are no references. There is also no mention of such “hot” topics as data mining or object-oriented databases. That is how it should be. The book is of primary interest to practitioners who want to know how to set up a data warehouse right now, but researchers will find in the book a good indication of the real needs of the business community. Data processing has two aspects: online processing of incoming data in a transaction database, and the analysis of historical data in a data warehouse in support of management decision making. The thesis of this book is that the two kinds of activities are quite distinct, that the transaction database and the data warehouse should be separate, and that different software systems are needed to manage them. The author expands on this main theme by examining several very large data warehouses he has himself designed for client companies (10 of 17 chapters). He then looks at the administrative side (5 chapters). In the final chapter, he takes a brief look into the future, but this is restricted to technological advances. Throughout the text, new design principles are introduced, which are then brought together in an appendix. Another appendix holds a detailed outline of the process for implementing a data warehouse. A CD-ROM supports the text. It contains sample data relating to the case studies, and software for managing the warehouse.