Relational database performance tuning was probably a topic of discussion even before commercial relational databases were released some 30 years ago. In the past decade, database tuning has been the subject of numerous academic articles and books, many of which focus on tuning techniques for specific commercial database products. The two key title words that set this book apart are “automated” and “physical.” This work brings the reader up to date on the body of research focused on automating the design and tuning of the physical layer of relational database systems. The author states that the book should be valuable to database professionals in academia and industry, and also to experienced database administrators (DBAs). Although DBAs may find the discussion of existing tuning tools useful, the bulk of the material will primarily be of interest to database researchers focusing on physical layer database performance.

The book is divided into three logical sections. Part 1 introduces the technical background for the material to follow: structured query language (SQL), query optimization techniques, indexes, and the complexity of the physical design problem. Although much of the early material will be familiar to the knowledgeable reader, the exposition is clear and compact, and leads to a rigorous definition of the physical design problem that is shown to be nondeterministic polynomial-time (NP) complete. Furthermore, the book shows the problem to be difficult to solve manually, even for moderately sized examples, hence motivating the need for efficient automated search techniques. The section concludes by introducing a useful cost model for evaluating a workload under a given index configuration.

Part 2 contains the key ideas and results. The discussion begins with automated index selection for a single query, later expanding to a generalized workload. The author clearly demonstrates that the optimal set of indexes for a generalized workload may not be a part of the optimal candidate set of indexes for any query in the workload. The discussion consistently focuses on the search space of alternative sets of indexes, the importance of a cost model for comparing alternatives, and efficient traversal of the search space. The section ends with a practical discussion of how one gathers and stores workload information, and introduces a real-world tuning tool for use with Microsoft SQL Server 2005 and above.

The final section of the book focuses on contemporary challenges to automated physical design, including materialized views, horizontal and vertical partitioning, and multidimensional clustering. Although the design and tuning problems are only compounded by these options, the author nicely extends the cost model and search space paradigm developed in prior chapters to explore these additional possibilities.

Although many of the results in this book were previously published in research articles from members of the Data Management, Exploration and Mining (DMX) group at Microsoft Research, the book is more than an edited collection of research papers. The writing is consistently good, the approach is mathematically sound, and the content will be valuable to anyone interested in physical database tuning, either manual or automated.