Browse Books

Reviewer: Alan Raymond Hevner

This book covers in detail the theory and techniques for providing concurrency control and recovery in database systems. The book is divided into eight chapters. Chapter 1 motivates the concurrency control and recovery problems. A database system model is presented and the serializability theory is introduced as a means for evaluating the correctness of concurrent executions of database transactions. Chapter 2 expands on the serializability theory. This theory provides a rigorous foundation for developing concurrency control and recovery algorithms. Chapter 3 covers the most popular method of concurrency control, two phase locking. Variations of two phase locking algorithms are presented, and the correctness and performance of each are analyzed. The management of deadlocks is an important issue. Chapter 4 surveys concurrency control methods that do not use locking. Such methods include timestamp ordering, graph testing, and certifiers (i.e., optimistic methods). While these methods are less popular than two phase locking, they do provide advantages in certain database system environments. Chapter 5 discusses concurrency control on multiversion database systems. Several versions of the same data are maintained for historical or recovery purposes. Chapter 6 presents recovery algorithms for centralized database systems. Recovery techniques that are surveyed include undo-redo logging, deferred updating, shadowing, checkpointing, and archiving. A framework for comparing the different algorithms based on their undo-redo characteristics is given. Chapter 7 extends recovery into distributed system environments. The critical problem of atomic commitment is studied. A transaction's results must be committed at every system site or at none of them. Ensuring atomic commitment in the presence of site and communication failures is a difficult problem. Algorithms for distributed recovery are presented and analyzed. Chapter 8 treats the concurrency control and recovery problems for replicated, distributed data. Algorithms for managing replicated data integrate concurrency control and recovery processing. Each chapter ends with bibliographic notes and a set of demanding exercises. An excellent glossary and bibliography, with citations through 1986, complete the book. This book will appeal to both database system researchers and developers. It brings together a great amount of research in a coherent treatment. The authors have done an excellent job of providing frameworks in which the different methods of concurrency control and recovery can be compared and analyzed. The tone of the book throughout is quite rigorous. Nevertheless, developers will find accessible presentations of the most important algorithms and accompanying discussions of performance and feasibility in various database system environments.