A Unified Approach to Concurrency Control and Transaction Recovery (Extended Abstract)

Conventional transaction management in shared databases is traditionally viewed as a combination of two orthogonal problems. One is to guarantee correctness when several transactions are executed concurrently. This is ensured by the concurrency control component of a transaction processing system. The other is to ensure the correctness of the database in spite of transaction aborts and system failures. This is ensured by the recovery mcchanism. In both cases there are several criteria that impose increasing restrictions on executions, resulting in a hierarchy in which only the most restrictive criteria are chosen for implementing concurrency control and recovery mechanisms. The result is that the different criteria are incomparable except for the most restrictive criterion in each component. For instance, strict two phase locking and log based recovery with before images has become the de facto standard (i.e. rigorous schedules [BGRS91], once considered as a possible candidate for a unified treatment of concurrency control and transaction recovery). Recently, Schek, Weikum and Ye [SWY93] have developed a unified theory for concurrency control and recovery. In this theory, the traditional concurrency control theory is expanded to include the actions of aborted transactions. As a result, the extended theory can be used to determine whether a given execution is correct both from a concurrency control as well as recovery points of view. A hierarchy of correctness classes is defined and the authors introduce a particular class, called prefix-reducibility (PRED). PRED is argued to be the most