Extending locking techniques to improve concurrent database access

Although the general concepts provided by the standard concurrency control theory (e. g. [BHG87]) remain a solid guide for systems behavior and design, a look at implementation-driven research (e. g. ARIES/IM [ML92]) shows that there is a nontrivial gap between the usual abstract model and practical DBMS situations. For example, most theoretical work on concurrency control has ignored access to redundant data stores, which are the basis for indices; and even simple operations such as insertion and deletion of records introduce unexpected subtleties in their correct concurrent handling. In this paper we propose extensions to the concepts of standard concurrency control to cover locking problems found in practice. Our results should help generalize previous proposals, understand the tradeo s involved, and facilitate the development of robust implementations.